ENCODING AND DECODING METHOD AND ELECTRONIC DEVICE

Description

1 / 79 “ENCODING AND DECODING METHOD AND ELECTRONIC DEVICE” TECHNICAL FIELD

[0001] The embodiments of the present invention relate to the field of encoding and decoding and, in particular, to encoding and decoding methods, and to an electronic device. BACKGROUND

[0002] An artificial intelligence (AI) image compression algorithm is implemented based on deep learning, and has a better compression effect than conventional image compression technology (such as Joint Photographic Experts Group, JPEG, or Better Portable Graphics, BPG). A deep learning-based end-to-end image encoding solution is widely used.

[0003] Currently, in an existing end-to-end image encoding process, a codec skips encoding and decoding entropy, which can introduce an artifact. SUMMARY

[0004] With this in mind, the present invention provides encoding and decoding methods, and an electronic device. The encoding and decoding methods can effectively prevent the introduction of an artifact into a reconstructed image.

[0005] According to a first aspect, an embodiment of the present invention provides a coding method. The method includes: first, obtaining an image to be coded; then, extracting a feature map from the image to be coded, and determining a probability estimation result of a region to be coded in the feature map; then, determining, based on an original feature value of the region to be coded and a reconstructed feature value of the region to be coded that is determined after adaptive entropy coding skip processing is performed on the region to be coded, whether adaptive entropy coding skip processing is performed on the region to be coded; if so, performing adaptive entropy coding skip processing. Petition 870250082415, dated 12 / 09 / 2025, page 6 / 95 2 / 79 entropy in at least one sub-block in the region to be encoded or skip entropy encoding processing in at least one sub-block based on the probability estimation result of the region to be encoded, and set a flag that matches the region to be encoded to a first value; or if not, perform entropy encoding processing in the region to be encoded based on the probability estimation result of the region to be encoded, and set a flag that matches the region to be encoded to a second value; and then encode the flag.

[0006] If the difference between the original feature value of the region to be encoded and the reconstructed feature value of the region to be encoded, which is determined after adaptive entropy coding skip processing is performed on the region to be encoded, is small, this indicates that performing adaptive entropy coding skip processing on the region to be encoded does not introduce an artifact in the reconstructed image of the region to be encoded. In this case, adaptive entropy coding skip processing can be performed on the region to be encoded.If the difference between the original feature value of the region to be encoded and the reconstructed feature value of the region to be encoded, determined after adaptive entropy coding skip processing is performed on the region to be encoded, is large, this indicates that performing adaptive entropy coding skip processing on the region to be encoded introduces an artifact into the reconstructed image of the region to be encoded. In this case, adaptive entropy coding skip processing may not be performed on the region to be encoded. This can effectively avoid the introduction of the artifact into the reconstructed image.

[0007] In addition, the flag is introduced to represent whether adaptive entropy coding skip processing is performed in the region to be encoded, and the flag is written to a bitstream. In this way, the flag can be transferred to a decoder, so that the decoder learns whether it performs adaptive entropy decoding skip processing in a corresponding region to be decoded.

[0008] For example, the feature map of the image to be encoded can be extracted using the feature extraction module. The feature extraction module can be implemented based on Petition 870250082415, dated 12 / 09 / 2025, page 7 / 95 3 / 79 a neural network. For example, the feature extraction module can be implemented based on a convolutional neural network (CNN). The CNN can be a neural network including a convolutional layer, and the neural network can additionally include a network layer, such as an activation layer (such as ReLU or PReLU), a pooling layer, a batch normalization layer (BN layer), or a fully connected layer. Conventional convolutional neural networks include LeNet, AlexNet, VGGNet, ResNet, and similar networks. For example, a basic CNN can include a backbone network and a head network; and a complex CNN includes a backbone network, a neck network, and a head network. A network included in the CNN is not limited in the present invention.For example, the feature extraction module can be implemented through multi-layer convolution stacking and combination with a nonlinear mapping activation function.

[0009] For example, the feature map is three-dimensional data that is emitted by the convolutional layer, the activation layer, the clustering layer, the batch normalization layer, and similar layers in the convolutional neural network, and the three dimensions are referred to, respectively, as a width, a height, and a channel. For example, the feature map might include W*H*C feature points, and each feature point has a corresponding feature value. For example, the feature value might be a floating-point number, where W, H, and C are all positive integers.

[0010] For example, a feature map used for entropy encoding processing and adaptive entropy encoding jump processing can be a quantized feature map.

[0011] For example, the feature map can be divided into a plurality of cubes using w1xh1xc1 as a basic unit (w1 is an integer in the range of 1 to W, h1 is an integer in the range of 1 to H, and c1 is an integer in the range of 1 to C). When c1=1, a cube means a block. A cube can be used as a region to be encoded.

[0012] In one possible mode, an estimation result of Petition 870250082415, dated 12 / 09 / 2025, page 8 / 95 4 / 79 The probability of each characteristic point can be a probability P.

[0013] In one possible mode, when a probability estimation module performs Gaussian distribution estimation, a probability estimation result may include a Gaussian distribution parameter: a mean μ and a variance σ.

[0014] For example, the region to be encoded may include a plurality of feature points, and at least one feature point may be grouped into a sub-block. In this way, the region to be encoded may include a plurality of sub-blocks, and each sub-block may include at least one feature point. Adaptive entropy coding skip processing can be: perform entropy coding processing on at least one sub-block in the region to be encoded, or skip entropy coding processing on at least one sub-block based on the probability estimation result of the region to be encoded. In this way, adaptive entropy coding skip processing can be performed on the region to be encoded using a sub-block as a minimal unit, thereby improving coding efficiency and reducing bit rate overheads.

[0015] For example, when a subblock includes a feature point, adaptive entropy coding skip processing can be: perform entropy coding processing on at least one feature point in the region to be coded, or skip entropy coding processing on at least one feature point based on the probability estimation result of the region to be coded. In this way, the region to be coded can be coded using a feature point as a minimum unit. When compared to coding the region to be coded using a subblock as a minimum unit, this can improve reconstruction quality.

[0016] In one possible mode, “setting a flag that corresponds to the region to be encoded for a first value” can be performed first, then “encoding the flag” is performed, and then “performing entropy encoding processing on at least one sub-block in the region to be encoded or skipping entropy encoding processing on at least one sub-block based on the probability estimation result of the region to be encoded” Petition 870250082415, dated 12 / 09 / 2025, page 9 / 95 5 / 79 is completed.

[0017] In one possible mode, “perform entropy encoding processing on at least one sub-block in the region to be encoded or skip entropy encoding processing on at least one sub-block based on the probability estimation result of the region to be encoded” can be performed first, then “set a flag that matches the region to be encoded to a first value” is performed, and then “encode the flag” is performed.

[0018] In one possible mode, “setting a flag that corresponds to the region to be encoded for a second value” can be performed first, then “encoding the flag” is performed, and then “performing entropy encoding processing on the region to be encoded based on the probability estimation result of the region to be encoded” is performed.

[0019] In one possible mode, “perform entropy encoding processing on the region to be encoded based on the probability estimation result of the region to be encoded” can be performed first, then “set a flag that matches the region to be encoded to a second value” is performed, and then “encode the flag” is performed.

[0020] In other words, the present invention does not limit a flag encoding sequence and perform entropy encoding processing or adaptive entropy encoding jump processing in the region to be encoded.

[0021] For example, after entropy encoding processing or adaptive entropy encoding jump processing is performed, a bitstream of the region to be encoded can be obtained; bitstreams of all regions to be encoded in the feature map can form a bitstream of the image to be encoded; and the flag can be written to the bitstream as a new syntax element.

[0022] For example, the flag can be encoded to obtain a bitstream of the flag; and bitstreams of all regions to be encoded in the feature map and bitstreams of corresponding flags can form a bitstream of the image to be encoded.

[0023] It should be understood that there may be a plurality of flag encoding modes. This is not limited to the present invention. Petition 870250082415, dated 12 / 09 / 2025, page 10 / 95 6 / 79

[0024] For example, the flag can be represented by a jump flag, a flag, or a cube flag.

[0025] For example, the first value can be set as required, for example, 1. The second value can be set as required, for example, 0. The first value is different from the second value.

[0026] According to the first aspect, coding the flag includes: determining a flag value probability based on the probability estimation result of the region to be coded, where the flag value probability is a probability that the flag is the first value or a probability that the flag is the second value; and coding the flag based on the flag value probability.

[0027] For example, the flag can alternatively be written to the bitstream after flag binarization. When compared to writing the flag to the bitstream after flag binarization, encoding the flag based on the flag value probability can reduce bit rate overheads.

[0028] In this way, the flag is encoded using the probability estimation result that can be obtained by both an encoder and a decoder, so that the decoder decodes the bit stream to obtain the flag.

[0029] According to the first aspect or any of the preceding implementations of the first aspect, determining, based on the original feature value of the region to be encoded and the reconstructed feature value of the region to be encoded that is determined after adaptive entropy coding jump processing is performed on the region to be encoded, whether to perform adaptive entropy coding jump processing on the region to be encoded includes: determining an error between the reconstructed feature value and the original feature value; and when the error is greater than an error threshold, determining to perform adaptive entropy coding jump processing on the region to be encoded; or when the error is less than or equal to an error threshold, determining not to perform adaptive entropy coding jump processing on the region to be encoded.

[0030] For example, the original feature value of the region to be encoded may include original feature values ​​of the plurality of Petition 870250082415, dated 12 / 09 / 2025, page 11 / 95 7 / 79 feature points are included in the region to be encoded, and the reconstructed feature value of the region to be encoded may include reconstructed feature values ​​from the plurality of feature points included in the region to be encoded.

[0031] For example, based on the error between the reconstructed feature value and the original feature value of the region to be encoded, it can be determined whether adaptive entropy coding skip processing is performed in the region to be encoded. For example, the error can be compared to the error threshold Thr. The error threshold Thr can be defined based on a maximum error between the reconstructed feature value and the original feature value of the region to be encoded when no artifact occurs in the reconstructed image of the region to be encoded. This is not limited to the present invention. When the error between the reconstructed feature value and the original feature value of the region to be encoded is greater than the error threshold Thr, this indicates that performing adaptive entropy coding skip processing in the region to be encoded introduces an artifact in the reconstructed image of the region to be encoded.In this case, adaptive entropy coding skip processing may not be performed in the region to be encoded. When the error between the reconstructed feature value and the original feature value of the region to be encoded is less than or equal to the error threshold Thr, it can be determined that performing adaptive entropy coding skip processing in the region to be encoded does not introduce an artifact in the reconstructed image of the region to be encoded. In this case, adaptive entropy coding skip processing can be performed in the region to be encoded.

[0032] According to the first aspect or any of the preceding implementations of the first aspect, the region to be encoded includes a plurality of subblocks, and the probability estimation result of the region to be encoded includes probability estimation results of the plurality of subblocks; and determining the flag value probability based on the probability estimation result of the region to be encoded includes: determining the flag value probability based on a sum of the probability estimation results of the plurality of subblocks in the region to be encoded. Petition 870250082415, dated 12 / 09 / 2025, p. 12 / 95 8 / 79

[0033] For example, when the probability estimation result is a Gaussian distribution parameter, Gaussian distribution parameters of all feature points included in all sub-blocks in the region to be encoded can be accumulated to obtain a Gaussian distribution parameter of the flag. A flag mean is μ = ^μ, and a flag variance is af = ^σ. Then, a probability of a Gaussian distribution with mean μ and variance σf in an interval of [0.5, 0.5] is used as the probability that the flag is the first value.

[0034] According to the first aspect or any of the preceding implementations of the first aspect, the region to be encoded includes a plurality of sub-blocks, the sub-block includes at least one feature point, and the probability estimation result of the region to be encoded includes probability estimation results of a plurality of feature points;Determining the flag value probability based on the probability estimation result of the region to be encoded includes: determining the flag value probability based on a number of feature points where entropy coding processing is skipped in the region to be encoded and a total number of feature points in the region to be encoded, where the number of feature points where entropy coding processing is skipped in the region to be encoded is determined based on the probability estimation results of the plurality of feature points included in the region to be encoded.

[0035] In one possible mode, when a probability P of a feature point value y[w][h][c] being an original feature value kl is greater than or equal to T0, the feature point y[w][h][c] is determined as a feature point at which entropy encoding processing is skipped.

[0036] In one possible mode, when a Gaussian distribution parameter of a characteristic point y[w][h][c] satisfies Petition 870250082415, dated 12 / 09 / 2025, page 13 / 95 9 / 79 abs(μ-k 1) + σ < T1, the feature point y [w][h][c] is determined as a feature point where entropy encoding processing is skipped.

[0037] In one possible mode, when a probability estimation result of a sub-block satisfies a predefined condition, a feature point y[w][h][c] in the sub-block is determined as a feature point at which entropy encoding processing is skipped.

[0038] In one possible mode, the predefined condition may be that a probability P of a subblock is greater than or equal to T2. The probability P of the subblock may be an average of probabilities P of all feature points in the subblock, or it may be a maximum or minimum value of probabilities P of all feature points in the subblock. This is not limited in the present invention. T2 may be a predefined limit. T2 may be any value between 0 and 1. For example, T2 is 0.99, 0.98, 0.97, or 0.95. This is not limited in the present invention. In one possible mode, T2 may be equal to T0. In one possible mode, T2 may not be equal to T0.

[0039] In one possible mode, when the probability estimation result is a Gaussian distribution parameter, the predefined condition may be that a Gaussian distribution parameter of a subblock satisfies abs(μ-k 1) + σ < T3 . In one possible mode, a subblock mean μ may be a mean of means μ of all feature points in the subblock, and a subblock variance σ may be a mean of variances σ of all feature points in the subblock. In one possible mode, a subblock mean μ may be a maximum or minimum value of means μ of all feature points in the subblock, and a subblock variance σ may be a maximum or minimum value of variances σ of all feature points in the subblock. T3 may be a predefined threshold. T3 can be any value between 0 and 1. For example, T3 is 0.2, 0.3, or 0.4. This is not limited in the present invention. In one possible embodiment, T3 can be equal to T1.In one possible scenario, T3 may not be equal to T1.

[0040] According to the first aspect or any of the preceding implementations of the first aspect, the region to be encoded Petition 870250082415, dated 12 / 09 / 2025, p. 14 / 95 10 / 79 includes a plurality of sub-blocks, the probability estimation result of the region to be encoded includes probability estimation results of the plurality of sub-blocks, the sub-block includes at least one feature point, and the probability estimation result of the sub-block includes a probability estimation result of a feature point included in the sub-block; and perform entropy coding processing on at least one sub-block in the region to be encoded or skip entropy coding processing on at least one sub-block based on the probability estimation result of the region to be encoded includes: for a first sub-block in the plurality of sub-blocks: when a probability estimation result of the first sub-block satisfies a predefined condition, skip entropy coding processing on all feature points in the first sub-block;Or, when a probability estimation result from the first sub-block does not satisfy a predefined condition, perform entropy encoding processing on all feature points in the first sub-block based on probability estimation results from all feature points in the first sub-block.

[0041] According to a second aspect, an embodiment of the present invention provides a decoding method. The method includes: first receiving a bit stream; then, determining a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream, and decoding the bit stream to obtain a flag that corresponds to the region to be decoded; when the flag is a first value, performing entropy decoding processing on at least one sub-block in the region to be decoded or setting a feature value of a feature point in at least one sub-block to a specific value based on the probability estimation result of the region to be decoded;Or, when the flag is a second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and then perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image. Since the flag is defined by an encoder; Petition 870250082415, dated 12 / 09 / 2025, page 15 / 95 Based on whether adaptive entropy coding skip processing causes an artifact to be introduced into the reconstructed image, a decoder determines, based on the flag in a decoding process, whether adaptive entropy coding skip processing or entropy coding skip processing is performed in the region to be decoded. This can prevent the introduction of an artifact into a reconstructed image obtained through decoding.

[0042] According to the second aspect, decoding the bit stream to obtain the flag that corresponds to the region to be decoded includes: determining a probability of the flag value based on the probability estimation result of the region to be decoded, where the probability of the flag value is a probability that the flag is the first value or a probability that the flag is the second value; and decoding the bit stream to obtain the flag that corresponds to the region to be decoded based on the probability of the flag value.

[0043] According to the second aspect or any of the preceding implementations of the second aspect, the region to be decoded includes a plurality of subblocks, and the probability estimation result of the region to be decoded includes probability estimation results of the plurality of subblocks; and determining the flag value probability based on the probability estimation result of the region to be decoded includes: determining the flag value probability based on a sum of the probability estimation results of the plurality of subblocks in the region to be decoded.

[0044] According to the second aspect or any of the preceding implementations of the second aspect, the region to be decoded includes a plurality of sub-blocks, the sub-block includes at least one feature point, and the probability estimation result of the region to be decoded includes probability estimation results of a plurality of feature points; and determining the flag value probability based on the probability estimation result of the region to be decoded includes: determining, based on the probability estimation results of the plurality of feature points included in the region to be decoded, a number of feature points on which processing of Petition 870250082415, dated 12 / 09 / 2025, p. 16 / 95 12 / 79 entropy decoding is skipped in the region to be decoded; and determine the probability of the flag value based on the number of feature points at which entropy decoding processing is skipped in the region to be decoded and a total number of feature points in the region to be decoded.

[0045] According to the second aspect or any of the preceding implementations of the second aspect, the region to be decoded includes a plurality of sub-blocks, the probability estimation result of the region to be decoded includes probability estimation results of the plurality of sub-blocks, the sub-block includes at least one feature point, and the probability estimation result of the sub-block includes a probability estimation result of a feature point included in the sub-block;and perform entropy decoding processing on at least one subblock in the region to be decoded, or set the feature value of the feature point in at least one subblock to a specific value based on the probability estimation result of the region to be decoded, to obtain the feature value of the feature point in the region to be decoded includes: for a first subblock in a plurality of subblocks: when a probability estimation result of the first subblock satisfies a predefined condition, set feature values ​​of all feature points in the first subblock to a specific value;Or, when a probability estimation result from the first sub-block does not satisfy a predefined condition, perform entropy decoding processing on all feature points in the first sub-block based on probability estimation results from all feature points in the first sub-block, to obtain feature values ​​from all feature points in the first sub-block.

[0046] The second aspect and any of the implementations of the second aspect correspond, respectively, to the first aspect and any of the implementations of the first aspect. For the technical effect that corresponds to the second aspect and any of the implementations of the second aspect, see the technical effect that corresponds to the first aspect and any of the implementations of the first aspect. The details are not described again in the present invention. Petition 870250082415, dated 12 / 09 / 2025, p. 17 / 95 13 / 79

[0047] According to a third aspect, the present invention further provides a coding method. The coding method includes: first, obtaining an image to be coded; then, extracting a feature map from the image to be coded, and determining a probability estimation result of a region to be coded in the feature map; then, determining, based on an original feature value of the region to be coded and a reconstructed feature value of the region to be coded that is determined after entropy coding processing in the region to be coded is skipped, whether to skip entropy coding processing in the region to be coded; if so, skipping entropy coding processing in the region to be coded, and setting a flag that corresponds to the region to be coded for a first value;Or, if not, perform entropy coding processing on the region to be coded based on the probability estimation result of the region to be coded, and set a flag that corresponds to the region to be coded for a second value; then, determine a flag value probability based on the probability estimation result of the region to be coded, where the flag value probability is either a probability that the flag is the first value or a probability that the flag is the second value; and encode the flag based on the flag value probability.

[0048] If the difference between the original feature value of the region to be encoded and the reconstructed feature value of the region to be encoded, which is determined after entropy coding processing in the region to be encoded is skipped, is small, this indicates that skipping entropy coding processing in the region to be encoded does not introduce an artifact in a reconstructed image of the region to be encoded. In this case, entropy coding processing in the region to be encoded can be skipped. If the difference between the original feature value of the region to be encoded and the reconstructed feature value of the region to be encoded, which is determined after entropy coding processing in the region to be encoded is skipped, is large, this indicates that skipping entropy coding processing in the region to be encoded introduces an artifact in a reconstructed image of the region to be encoded. In this case, Petition 870250082415, dated 12 / 09 / 2025, page 18 / 95 14 / 79 Entropy encoding processing in the region to be encoded may not be skipped. This can effectively prevent the introduction of the artifact into the reconstructed image.

[0049] Furthermore, skipping entropy encoding processing across the entire region to be encoded can improve entropy encoding skip processing efficiency, thereby improving encoding efficiency.

[0050] Furthermore, when compared to writing the flag in a bitstream after flag binarization, encoding the flag based on the probability of the flag value can reduce bit rate overheads. The flag is encoded using the probability estimation result that can be obtained by both an encoder and a decoder, such that the decoder decodes the bitstream to obtain the flag.

[0051] According to the third aspect, determine, based on the original feature value of the region to be encoded and the reconstructed feature value of the region to be encoded that is determined after entropy coding processing in the region to be encoded is skipped, whether skipping entropy coding processing in the region to be encoded includes: determining an error between the reconstructed feature value and the original feature value; and when the error is greater than an error threshold, determine whether to skip entropy coding processing in the region to be encoded; or when the error is less than or equal to an error threshold, determine whether not to skip entropy coding processing in the region to be encoded.

[0052] According to the third aspect or any of the preceding implementations of the third aspect, the region to be encoded includes a plurality of feature points, and the probability estimation result of the region to be encoded includes probability estimation results of the plurality of feature points in the region to be encoded; and determining the flag value probability based on the probability estimation result of the region to be encoded includes: determining the flag value probability based on a sum of the probability estimation results of the plurality of feature points in the region to be encoded. Petition 870250082415, dated 12 / 09 / 2025, p. 19 / 95 15 / 79

[0053] According to the third aspect or any of the preceding implementations of the third aspect, the region to be encoded includes a plurality of feature points, and the probability estimation result of the region to be encoded includes probability estimation results of the plurality of feature points in the region to be encoded; and determining the flag value probability based on the probability estimation result of the region to be encoded includes: determining, based on the probability estimation result of the region to be encoded, a number of feature points at which entropy encoding processing is skipped in the region to be encoded; and determining the flag value probability based on the number of feature points at which entropy encoding processing is skipped in the region to be encoded and a total number of feature points in the region to be encoded.

[0054] According to the third aspect or any of the preceding implementations of the third aspect, determining, based on the probability estimation result of the region to be encoded, the number of feature points at which entropy encoding processing is skipped in the region to be encoded includes: Use a number of feature points whose probability estimation results satisfy a predefined condition in the region to be encoded as the number of feature points at which entropy encoding processing is skipped in the region to be encoded.

[0055] The third aspect and any of the implementations of the third aspect correspond, respectively, to the first aspect and any of the implementations of the first aspect. For the technical effect that corresponds to the third aspect and any of the implementations of the third aspect, see the technical effect that corresponds to the first aspect and any of the implementations of the first aspect. The details are not described again in the present invention.

[0056] According to a fourth aspect, the present invention further provides a decoding method. The method includes: first, receiving a bit stream; then, determining a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream; then, determining a probability value. Petition 870250082415, dated 12 / 09 / 2025, p. 20 / 95 16 / 79 of a flag that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded, where the flag value probability is either a probability that the flag is a first value or a probability that the flag is a second value; decode the bitstream to obtain the flag based on the flag value probability; when the flag is the first value, set a feature value of a feature point in the region to be decoded to a specific value; or when the flag is the second value, perform entropy decoding processing in the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded;and then perform reconstruction based on a feature value from a feature point on the feature map, to obtain a reconstructed image.

[0057] According to the fourth aspect, the region to be decoded includes a plurality of feature points, and the probability estimation result of the region to be decoded includes probability estimation results of the plurality of feature points in the region to be decoded; and determining the probability value of the flag that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded includes: determining the probability value of the flag based on a sum of the probability estimation results of the plurality of feature points in the region to be decoded.

[0058] According to the fourth aspect or any of the preceding implementations of the fourth aspect, the region to be decoded includes a plurality of feature points, and the probability estimation result of the region to be decoded includes probability estimation results of the plurality of feature points in the region to be decoded; and determining the probability value of the flag that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded includes: determining, based on the probability estimation result of the region to be decoded, a number of feature points at which entropy decoding processing is skipped in the region to be decoded; and determining the probability value of the flag based on the number of feature points at which processing Petition 870250082415, dated 12 / 09 / 2025, p. 21 / 95 17 / 79 of entropy decoding is skipped in the region to be decoded and a total number of feature points in the region to be decoded.

[0059] According to the fourth aspect or any of the preceding implementations of the fourth aspect, determining, based on the probability estimation result of the region to be decoded, the number of feature points at which entropy decoding processing is skipped in the region to be decoded includes: using a number of feature points whose probability estimation results satisfy a predefined condition in the region to be decoded as the number of feature points at which entropy decoding processing is skipped in the region to be decoded.

[0060] The fourth aspect and any of the implementations of the fourth aspect correspond, respectively, to the first aspect and any of the implementations of the first aspect. For the technical effect that corresponds to the fourth aspect and any of the implementations of the fourth aspect, see the technical effect that corresponds to the first aspect and any of the implementations of the first aspect. The details are not described again in the present invention.

[0061] According to a fifth aspect, the present invention additionally provides an encoder. The encoder includes: a capture module, configured to obtain an image to be encoded; A feature extraction module, configured to extract a feature map from the image to be encoded; A probability estimation module, configured to determine a probability estimation result for a region to be encoded in the feature map; and an entropy encoding module, configured to: determine, based on an original feature value of the region to be encoded and a reconstructed feature value of the region to be encoded that is determined after adaptive entropy coding skip processing is performed on the region to be encoded, whether to perform adaptive entropy coding skip processing on the region to be encoded; if so, perform entropy coding processing on at least one sub-block. Petition 870250082415, dated 12 / 09 / 2025, page 22 / 95 18 / 79 in the region to be encoded or skip entropy encoding processing in at least one sub-block based on the probability estimation result of the region to be encoded, and set a flag that matches the region to be encoded to a first value; or otherwise, perform entropy encoding processing in the region to be encoded based on the probability estimation result of the region to be encoded, and set a flag that matches the region to be encoded to a second value; and encode the flag.

[0062] It should be understood that the coder may additionally perform the steps in the first aspect and any of the implementations of the first aspect.

[0063] The fifth aspect and any of the implementations of the fifth aspect correspond, respectively, to the first aspect and any of the implementations of the first aspect. For the technical effect that corresponds to the fifth aspect and any of the implementations of the fifth aspect, see the technical effect that corresponds to the first aspect and any of the implementations of the first aspect. The details are not described again in the present invention.

[0064] According to a sixth aspect, the present invention additionally provides a decoder. The decoder includes: a receiver module, configured to receive a bit stream; An entropy decoding module, configured to: determine a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream, and decode the bit stream to obtain a flag that corresponds to the region to be decoded; and when the flag is a first value, perform entropy decoding processing on at least one sub-block in the region to be decoded or set a feature value of a feature point in at least one sub-block to a specific value based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded;or when the flag is a second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and; Petition 870250082415, dated 12 / 09 / 2025, page 23 / 95 19 / 79 is a feature decoding module, configured to perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

[0065] It should be understood that the decoder can additionally perform the steps in the second aspect and any of the implementations of the second aspect.

[0066] The sixth aspect and any of the implementations of the sixth aspect correspond, respectively, to the first aspect and any of the implementations of the first aspect. For the technical effect that corresponds to the sixth aspect and any of the implementations of the sixth aspect, see the technical effect that corresponds to the first aspect and any of the implementations of the first aspect. The details are not described again in the present invention.

[0067] According to a seventh aspect, the present invention additionally provides an encoder. The encoder includes: a capture module, configured to obtain an image to be encoded; A feature extraction module, configured to extract a feature map from the image to be encoded; a probability estimation module, configured to determine a probability estimation result for a region to be encoded in the feature map; and an entropy encoding module, configured to: determine, based on an original feature value of the region to be encoded and a reconstructed feature value of the region to be encoded that is determined after entropy encoding processing on the region to be encoded is skipped, whether to skip entropy encoding processing on the region to be encoded; if so, skip entropy encoding processing on the region to be encoded, and set a flag that matches the region to be encoded to a first value;or, if not, perform entropy encoding processing on the region to be encoded based on the probability estimation result of the region to be encoded, and set a flag that corresponds to the region to be encoded for a second value; determine a probability value for the flag based on the result of; Petition 870250082415, dated 12 / 09 / 2025, page 24 / 95 20 / 79 probability estimation of the region to be encoded, where the flag value probability is either the probability that the flag is the first value or the probability that the flag is the second value; and encode the flag based on the flag value probability.

[0068] It should be understood that the coder may additionally perform the steps in the third aspect and any of the implementations of the third aspect.

[0069] The seventh aspect and any of the implementations of the seventh aspect correspond, respectively, to the first aspect and any of the implementations of the first aspect. For the technical effect that corresponds to the seventh aspect and any of the implementations of the seventh aspect, see the technical effect that corresponds to the first aspect and any of the implementations of the first aspect. The details are not described again in the present invention.

[0070] According to an eighth aspect, the present invention further provides a decoder. The decoder includes: a receiver module, configured to receive a bit stream; An entropy decoding module, configured to: determine a probability estimation result of a region to be decoded in a feature map that corresponds to the bitstream; determine a flag value probability that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded, where the flag value probability is either a probability that the flag is a first value or a probability that the flag is a second value; decode the bitstream to obtain the flag based on the flag value probability; and when the flag is the first value, set a feature value of a feature point in the region to be decoded to a specific value;or when the flag is the second value, perform entropy decoding processing in the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and a feature decoding module, configured to perform reconstruction based on a feature value of a feature point in the feature map, to obtain an image; Petition 870250082415, dated 12 / 09 / 2025, page 25 / 95 21 / 79 rebuilt.

[0071] It should be understood that the decoder can additionally perform the steps in the fourth aspect and any of the implementations of the fourth aspect.

[0072] The eighth aspect and any of the implementations of the eighth aspect correspond, respectively, to the first aspect and any of the implementations of the first aspect. For the technical effect that corresponds to the eighth aspect and any of the implementations of the eighth aspect, see the technical effect that corresponds to the first aspect and any of the implementations of the first aspect. The details are not described again in the present invention.

[0073] According to a ninth aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor. The memory is coupled to the processor. The memory stores program instructions and, when the program instructions are executed by the processor, the electronic device is enabled to perform the steps in the first aspect (or the third aspect) or any possible implementation of the first aspect (or the third aspect).

[0074] The ninth aspect and any of the implementations of the ninth aspect correspond, respectively, to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). For the technical effect that corresponds to the ninth aspect and any of the implementations of the ninth aspect, see the technical effect that corresponds to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). Details are not described again in the present invention.

[0075] According to a tenth aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor. The memory is coupled to the processor. The memory stores program instructions and when the program instructions are executed by the processor, the electronic device is enabled to perform the steps in the second aspect (or the fourth aspect) or any possible implementation of the second aspect (or the fourth aspect).

[0076] The tenth aspect and any of the implementations Petition 870250082415, dated 12 / 09 / 2025, page 26 / 95 22 / 79 of the tenth aspect correspond, respectively, to the second aspect (or the fourth aspect) and any of the implementations of the second aspect (or the fourth aspect). For the technical effect that corresponds to the tenth aspect and any of the implementations of the tenth aspect, see the technical effect that corresponds to the second aspect (or the fourth aspect) and any of the implementations of the second aspect (or the fourth aspect). The details are not described again in the present invention.

[0077] According to an eleventh aspect, an embodiment of the present invention provides a chip, including one or more interface circuits and one or more processors. The one or more processors receive or send data through the one or more interface circuits, and when the one or more processors execute computer instructions, an electronic device is enabled to perform the steps in the first aspect (or the third aspect) or any possible implementation of the first aspect (or the third aspect).

[0078] The eleventh aspect and any of the implementations of the eleventh aspect correspond, respectively, to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). For the technical effect corresponding to the eleventh aspect and any of the implementations of the eleventh aspect, see the technical effect corresponding to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). Details are not described again in the present invention.

[0079] According to a twelfth aspect, an embodiment of the present invention provides a chip, including one or more interface circuits and one or more processors. The one or more processors receive or send data through the one or more interface circuits, and when the one or more processors execute computer instructions, an electronic device is enabled to perform the steps in the second aspect (or the fourth aspect) or any possible implementation of the second aspect (or the fourth aspect).

[0080] The twelfth aspect and any of the implementations of the twelfth aspect correspond, respectively, Petition 870250082415, dated 12 / 09 / 2025, p. 27 / 95 23 / 79 to the second aspect (or to the fourth aspect) and any of the implementations of the second aspect (or of the fourth aspect). For the technical effect corresponding to the twelfth aspect and any of the implementations of the twelfth aspect, see the technical effect corresponding to the second aspect (or to the fourth aspect) and any of the implementations of the second aspect (or of the fourth aspect). The details are not described again in the present invention.

[0081] According to a thirteenth aspect, an embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium stores a computer program and, when the computer program is run on a computer or processor, the computer or processor is enabled to perform the steps in the first aspect (or the third aspect) or any possible implementation of the first aspect (or the third aspect).

[0082] The thirteenth aspect and any of the implementations of the thirteenth aspect correspond, respectively, to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). For the technical effect corresponding to the thirteenth aspect and any of the implementations of the thirteenth aspect, see the technical effect corresponding to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). Details are not described again in the present invention.

[0083] According to a fourteenth aspect, an embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium stores a computer program and, when the computer program is run on a computer or processor, the computer or processor is enabled to perform the steps in the second aspect (or the fourth aspect) or any possible implementation of the second aspect (or the fourth aspect).

[0084] The fourteenth aspect and any of the implementations of the fourteenth aspect correspond, respectively, to the second aspect (or the fourth aspect) and any of the implementations of the second aspect (or the fourth aspect). For technical purposes Petition 870250082415, dated 12 / 09 / 2025, p. 28 / 95 24 / 79 which corresponds to the fourteenth aspect and any of the implementations of the fourteenth aspect, see the technical effect which corresponds to the second aspect (or the fourth aspect) and any of the implementations of the second aspect (or the fourth aspect). The details are not described again in the present invention.

[0085] According to a fifteenth aspect, an embodiment of the present invention provides a computer program product. The computer program product includes computer instructions and, when the computer instructions are executed by a computer or a processor, the computer or processor is enabled to perform the steps in the first aspect (or the third aspect) or any possible implementation of the first aspect (or the third aspect).

[0086] The fifteenth aspect and any of the implementations of the fifteenth aspect correspond, respectively, to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). For the technical effect corresponding to the fifteenth aspect and any of the implementations of the fifteenth aspect, see the technical effect corresponding to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). Details are not described again in the present invention.

[0087] According to a sixteenth aspect, an embodiment of the present invention provides a computer program product. The computer program product includes computer instructions and, when the computer instructions are executed by a computer or a processor, the computer or processor is enabled to perform the steps in the second aspect (or the fourth aspect) or any possible implementation of the second aspect (or the fourth aspect).

[0088] The sixteenth aspect and any of the implementations of the sixteenth aspect correspond, respectively, to the second aspect (or the fourth aspect) and any of the implementations of the second aspect (or the fourth aspect). For the technical effect that corresponds to the sixteenth aspect and any of the implementations of the sixteenth aspect, see the technical effect that Petition 870250082415, dated 12 / 09 / 2025, p. 29 / 95 25 / 79 corresponds to the second aspect (or the fourth aspect) and any of the implementations of the second aspect (or the fourth aspect). The details are not described again in the present invention.

[0089] According to a seventeenth aspect, an embodiment of the present invention provides a bitstream storage apparatus. The apparatus includes a receiver and at least one storage medium. The receiver is configured to receive a bitstream. At least one storage medium is configured to store the bitstream. The bitstream is generated according to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect).

[0090] The seventeenth aspect and any of the implementations of the seventeenth aspect correspond, respectively, to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). For the technical effect corresponding to the seventeenth aspect and any of the implementations of the seventeenth aspect, see the technical effect corresponding to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). Details are not described again in the present invention.

[0091] According to an eighteenth aspect, an embodiment of the present invention provides a bitstream transmission apparatus. The apparatus includes a transmitter and at least one storage medium. The at least one storage medium is configured to store a bitstream. The bitstream is generated according to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). The transmitter is configured to obtain the bitstream from the storage medium, and send the bitstream to a device-side device via the transmission medium.

[0092] The eighteenth aspect and any of the implementations of the eighteenth aspect correspond, respectively, to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). For the technical effect that corresponds to the eighteenth aspect and any of the implementations of the eighteenth aspect, see the technical effect that Petition 870250082415, dated 12 / 09 / 2025, p. 30 / 95 26 / 79 corresponds to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). The details are not described again in the present invention.

[0093] According to a nineteenth aspect, an embodiment of the present invention provides a bitstream delivery system. The system includes: at least one storage medium, configured to store at least one bitstream, wherein the at least one bitstream is generated according to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect); and a media streaming device, configured to obtain a target bitstream from the at least one storage medium, and send the target bitstream to a device-side device, wherein the media streaming device includes a content server or a content delivery server.

[0094] The nineteenth aspect and any of the implementations of the nineteenth aspect correspond, respectively, to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). For the technical effect corresponding to the nineteenth aspect and any of the implementations of the nineteenth aspect, see the technical effect corresponding to the first aspect (or the third aspect) and any of the implementations of the first aspect (or the third aspect). Details are not described again in the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1a is a diagram of an exemplary application scenario;

[0096] Figure 1b is a diagram of an exemplary application scenario;

[0097] Figure 1c is a diagram of an exemplary end-to-end image compression process;

[0098] Figure 2 is a diagram of an exemplary coding process;

[0099] Figure 3 is a diagram of an exemplary decoding process;

[0100] Figure 4 is a diagram of a compression process of Petition 870250082415, dated 12 / 09 / 2025, page 31 / 95 27 / 79 image end to end illustrative;

[0101] Figure 5a is a diagram of an exemplary coding process;

[0102] Figure 5b is a diagram of an exemplary structure of an auxiliary information extraction module;

[0103] Figure 5c is a diagram of an exemplary structure of a probability estimation module;

[0104] Figure 6 is an example diagram of a decoding process;

[0105] Figure 7 is a diagram of an exemplary end-to-end image compression process;

[0106] Figure 8 is a diagram of an exemplary coding process;

[0107] Figure 9 is an example diagram of a decoding process;

[0108] Figure 10 is a diagram of an exemplary machine vision task based on encoding and decoding; and

[0109] Figure 11 is a diagram of an exemplary structure of a device. DETAILED DESCRIPTION

[0110] The following content clearly and completely describes the technical solutions in embodiments of the present invention with reference to the accompanying drawings in embodiments of the present invention. It is evident that the embodiments described are some, but not all, embodiments of the present invention. All other embodiments obtained by a person skilled in the art based on embodiments of the present invention without creative effort should be included within the scope of protection of the present invention.

[0111] An “and / or” term in this descriptive report describes only one association relationship between associated objects and indicates that three relationships can exist. For example, A and / or B can represent the following three cases: Only A exists, both A and B exist, and only B exists.

[0112] In this descriptive report and in the embodiments of the present invention, the terms “first”, “second” and similar are intended to distinguish between different objects, but do not indicate a particular order of Petition 870250082415, dated 12 / 09 / 2025, p. 32 / 95 28 / 79 objects. For example, a first target object, a second target object, and similar are used to distinguish between different target objects, but are not used to describe a specific order of the target objects.

[0113] In embodiments of the present invention, a word such as “example” or “for example” indicates an instance or description. Any embodiment or design solution described as an “example” or “for example” in embodiments of the present invention should not be explained as being more preferred or having more advantages than another embodiment or design solution. More precisely, the use of the word such as “example” or “for example” is intended to present a related concept in a specific way.

[0114] In descriptions of embodiments of the present invention, “a plurality of” means two or more, unless otherwise specified. For example, a plurality of processing units means two or more processing units, and a plurality of systems means two or more systems.

[0115] Figure 1a is a diagram of an exemplary application scenario. The application scenario in Figure 1a includes various video service scenarios based on video encoding and decoding, for example, a video conferencing scenario, a video call scenario, an online teaching scenario, a remote tutoring scenario, a low-latency live streaming scenario, a cloud gaming scenario, a wireless screen projection scenario, and a wireless extended screen scenario. This is not limited to embodiments of the present invention.

[0116] As shown in Figure 1a, for example, a first device may include a video capture module (e.g., a camera), a capture pre-processing module, an encoding module (or an encoder), and a transmission module. For example, the encoding module may be a software module, or it may be a hardware module. This is not limited to embodiments of the present invention. It should be understood that Figure 1a is only one example of the first device. In some other embodiments of the present invention, the first device has more modules than those shown in Figure 1a. This is not limited to embodiments of the present invention. Petition 870250082415, dated 12 / 09 / 2025, page 33 / 95 29 / 79

[0117] As shown in Figure 1a, for example, a second device may include a display module, a decoding module (or a decoder), and a receiver module. For example, the decoding module may be a software module, or it may be a hardware module. This is not limited to embodiments of the present invention. It should be understood that Figure 1a is only one example of the second device. In some other embodiments of the present invention, the second device has more modules than those shown in Figure 1a. This is not limited to embodiments of the present invention.

[0118] As shown in Figure 1a, for example, a process in which the camera of the first device captures original video data and then sends the captured original video data to the second device for display is as follows: The video capture module captures the original video data and then inserts the original video data into the capture preprocessing module. The capture preprocessing module performs a series of preprocessing operations on the original video data and outputs preprocessed original video data to the encoding module. The encoding module can then encode the preprocessed original video data to obtain a bitstream (the bitstream can also be referred to as a bitstream or a code stream), and output the bitstream obtained through encoding to the sending module. The sending module can then send the bitstream to the second device.Subsequently, the receiver module of the second device can receive the bitstream and then send the bitstream to the decoding module. The decoding module can then decode the bitstream to obtain reconstructed video data and send the reconstructed video data to the display module. The display module renders and displays the reconstructed video data.

[0119] For example, the first device includes, but is not limited to, a server, a personal computer (PC), a notebook computer, a tablet computer, a mobile phone and a watch.

[0120] For example, the second device includes, but is not limited to, a PC, a notebook computer, a tablet computer, a mobile phone and a watch.

[0121] For example, in the videoconferencing scenario, the first Petition 870250082415, dated 12 / 09 / 2025, page 34 / 95 30 / 79 device can be a PC or a notebook computer, and the second device can be a PC or a notebook computer.

[0122] For example, in an online teaching scenario, the first device might be a PC or a notebook computer, and the second device might be a tablet computer.

[0123] For example, in the cloud gaming scenario, the first device could be a server, and the second device could be a tablet computer, a PC, a notebook computer, or a mobile phone.

[0124] It should be understood that the present invention can alternatively be applied to an image encoding and decoding scenario. This is not limited to the present invention.

[0125] It should be understood that the sending module of the first device can alternatively send the bit stream directly to the receiving module of the second device without a network. This is not limited to the present invention.

[0126] It should be understood that the bit stream obtained through encoding can, alternatively, be stored directly.

[0127] Figure 1b is a diagram of an exemplary application scenario. The application scenario in Figure 1b is a computer vision task based on video (or image) encoding and decoding. The machine vision task includes, but is not limited to, object classification, recognition, segmentation, and the like. This is not limited to the present invention.

[0128] For a first device in Figure 1b, refer to the descriptions of the first device in Figure 1a. The details are not described again in the present invention.

[0129] As shown in Figure 1b, for example, a second device may include a receiver module, a decoding module (or a decoder), and a machine vision task module. The machine vision task module may be configured to perform a machine vision task.

[0130] As also shown in Figure 1b, for example, a process in which the camera of the first device captures video data. Petition 870250082415, dated 12 / 09 / 2025, page 35 / 95 The process of capturing the original video data (31 / 79) and then sending the captured original video data to the second device to perform a machine vision task is as follows: The video capture module captures the original video data and then inserts the original video data into the capture preprocessing module. The capture preprocessing module performs a series of preprocessing operations on the original video data and outputs the preprocessed original video data to the encoding module. The encoding module then encodes the preprocessed original video data to obtain a bitstream (the bitstream can also be referred to as a bitstream or a code stream), and outputs the resulting bitstream to the sending module. The sending module then sends the bitstream to the second device.Subsequently, the receiver module of the second device can receive the bit stream and then transmit the bit stream to the decoding module. Next, the decoding module can decode the bit stream and transmit a feature obtained in the decoding process to the machine vision task module. Then, the machine vision task module can execute the machine vision task based on the feature obtained in the decoding process, to obtain a machine vision task result.

[0131] For example, the first device includes, but is not limited to, a server, a personal computer (PC), a notebook computer, a tablet computer, a mobile phone and a watch.

[0132] For example, the second device includes, but is not limited to, a PC, a notebook computer, a tablet computer, a mobile phone and a watch.

[0133] It should be understood that the sending module of the first device can alternatively send the bit stream directly to the receiving module of the second device without a network. This is not limited to the present invention.

[0134] It should be understood that the present invention can alternatively be applied to another service scenario based on video (or image) encoding and decoding. This is not limited to the present invention.

[0135] Figure 1c is a diagram of an exemplary end-to-end image compression process. Petition 870250082415, dated 12 / 09 / 2025, page 36 / 95 32 / 79

[0136] As shown in Figure 1c, for example, an encoder may include a feature extraction module, a feature quantization module, a probability estimation module, and an entropy encoding module. It should be understood that the encoder may alternatively include more or fewer modules than those shown in Figure 1c. This is not limited to the present invention.

[0137] For example, the feature extraction module can be configured to perform feature extraction. For example, the feature quantization module can be configured to perform quantization. For example, the entropy coding module can be configured to perform entropy coding processing (e.g., Huffman Coding (HC), Arithmetic Coding (AC), and Asymmetric Numeral System (ANS)). For example, the probability estimation module can be configured to perform probability estimation.

[0138] As shown in Figure 1c, for example, a decoder may include a feature decoding module, a feature dequantization module, a probability estimation module, and an entropy decoding module. It should be understood that the decoder may alternatively include more or fewer modules than those shown in Figure 1c. This is not limited in the present invention. The feature dequantization module is an optional module.

[0139] For example, the entropy decoding module can be configured to perform entropy decoding processing. For example, the feature dequantization module can be configured to perform dequantization. For example, the feature decoding module can be configured to perform feature decoding to obtain a reconstructed image. For example, the probability estimation module can be configured to perform probability estimation.

[0140] Figure 2 is a diagram of an exemplary encoding process. Figure 2 shows the encoding process based on Figure 1a and Figure 1c. In the embodiment in Figure 2, an example of encoding an image is used for description. The image can be a standalone image, or it can be a frame in a video. This is not limited to Petition 870250082415, dated 12 / 09 / 2025, page 37 / 95 33 / 79 present invention. Steps S201 to S209 are all steps in the encoder coding process.

[0141] S201: Get an image to be encoded.

[0142] S202: Extract a feature map from the image to be encoded.

[0143] For example, after the encoder obtains the image to be encoded, the feature extraction module can extract a feature from the image to be encoded, in order to obtain the feature map of the image to be encoded.

[0144] For example, the feature extraction module can be implemented based on a neural network. For example, the feature extraction module can be implemented based on a convolutional neural network (CNN). The CNN can be a neural network including a convolutional layer, and the neural network can additionally include a network layer, such as an activation layer (such as ReLU or PReLU), a pooling layer, a batch normalization layer (BN layer), or a fully connected layer. Conventional convolutional neural networks include LeNet, AlexNet, VGGNet, ResNet, and the like. For example, a basic CNN can include a backbone network and a head network; and a complex CNN includes a backbone network, a neck network, and a head network. A network included in the CNN is not limited in the present invention.

[0145] For example, the feature extraction module can be implemented through multi-layer convolution stacking and combination with a nonlinear mapping activation function.

[0146] For example, a feature map is three-dimensional data emitted by the convolutional layer, the activation layer, the clustering layer, or the batch normalization layer in a convolutional neural network, and the three dimensions are referred to, respectively, as a width, a height, and a channel. For example, the feature map might include W*H*C feature points, and each feature point has a corresponding feature value. For example, the feature value might be Petition 870250082415, dated 12 / 09 / 2025, page 38 / 95 34 / 79 is a floating-point number, where W, H, and C are all positive integers.

[0147] So, the feature quantization module can quantize the feature map. It should be noted that the feature map in steps S203 to S209 can be a quantized feature map.

[0148] For example, the feature map may be divided into a plurality of regions; then, each region is determined as a region to be encoded sequentially; and then, steps S203 to S209 are performed for the region to be encoded, to encode the region to be encoded. The following content uses a region to be encoded as an example for description. The region to be encoded may be a two-dimensional region, or it may be a three-dimensional region. The region to be encoded may include one or more feature points. This is not limited in the present invention.

[0149] S203: Determine a probability estimation result for the region to be coded on the feature map.

[0150] For example, the probability estimation module can perform probability estimation at each feature point in the feature map, and determine a probability estimation result for each feature point in the feature map, so that the probability estimation result for the region to be encoded in the feature map can be obtained. The probability estimation result for the region to be encoded can include probability estimation results for all feature points in the region to be encoded. A probability estimation process is described subsequently.

[0151] S204: Determine, based on an original feature value of the region to be encoded and a reconstructed feature value of the region to be encoded that is determined after adaptive entropy coding skip processing is performed on the region to be encoded, whether adaptive entropy coding skip processing is performed on the region to be encoded.

[0152] For example, the reconstructed feature value of the region to be encoded can first be determined if adaptive entropy coding jump processing is performed on the region to be encoded; and, Petition 870250082415, dated 12 / 09 / 2025, page 39 / 95 35 / 79 then, the reconstructed feature value of the region to be encoded is compared to the original feature value to determine whether to perform adaptive entropy coding jump processing in the region to be encoded.

[0153] For example, it can be determined, by comparing the reconstructed feature value with the original feature value of the region to be encoded, whether an artifact is introduced after adaptive entropy coding skip processing is performed in the region to be encoded. When it is determined, based on a comparison result between the reconstructed feature value and the original feature value of the region to be encoded, that an artifact is introduced after adaptive entropy coding skip processing is performed in the region to be encoded, adaptive entropy coding skip processing may not be performed in the region to be encoded. In this case, steps S207, S208, and S209 can be performed.When it is determined, based on a comparison between the reconstructed feature value and the original feature value of the region to be encoded, that no artifact is introduced after adaptive entropy coding skip processing is performed on the region to be encoded, adaptive entropy coding skip processing can be performed on the region to be encoded. In this case, steps S205, S206, and S209 can be performed. A specific determination method is described subsequently. This can prevent the introduction of an artifact into a reconstructed image.

[0154] The original feature value of the region to be encoded may include original feature values ​​from a plurality of feature points included in the region to be encoded, and the reconstructed feature value of the region to be encoded may include reconstructed feature values ​​from the plurality of feature points included in the region to be encoded.

[0155] S205: Perform entropy coding processing on at least one subblock in the region to be coded or skip entropy coding processing on at least one subblock based on the probability estimation result of the region to be coded.

[0156] For example, the region to be encoded may include a Petition 870250082415, dated 12 / 09 / 2025, page 40 / 95 36 / 79 plurality of feature points, and at least one feature point can be grouped into a sub-block. In this way, the region to be encoded can be divided into a plurality of sub-blocks, and each sub-block can include at least one feature point. The probability estimation result of the region to be encoded can include probability estimation results of the plurality of sub-blocks, and the probability estimation result of each sub-block can include a probability estimation result of a feature point included in the sub-block. For example, adaptive entropy coding skip processing could be: perform entropy coding processing on at least one sub-block in the region to be encoded or skip entropy coding processing on at least one sub-block based on the probability estimation result of the region to be encoded.In this way, adaptive entropy coding skip processing can be performed in the region to be encoded using a sub-block as a minimal unit, thereby improving coding efficiency and reducing bit rate overheads.

[0157] Specifically, for each sub-block, it is determined, based on the sub-block probability estimation result, whether to skip entropy coding processing in the sub-block; and when it is determined not to skip entropy coding processing in the sub-block, entropy coding processing is performed on a feature value of a feature point in the sub-block based on a feature point probability estimation result in the sub-block; or when it is determined to skip entropy coding processing in the sub-block, entropy coding processing on all feature points in the sub-block is skipped; and after entropy coding processing on all feature points in the sub-block is skipped, feature values ​​of all feature points in the sub-block are set to a specific value (to facilitate determination of a reconstructed image of the region to be coded).

[0158] It should be understood that when the sub-block includes only one feature point, adaptive entropy coding jump processing is performed using one feature point as a minimum unit. In this case, entropy coding jump processing Petition 870250082415, dated 12 / 09 / 2025, page 41 / 95 37 / 79 adaptive encoding can be: performing entropy encoding processing on at least one feature point in the region to be encoded, or skipping entropy encoding processing on at least one feature point based on the probability estimation result of the region to be encoded. In this way, the region to be encoded can be encoded using a feature point as a minimum unit. When compared to encoding the region to be encoded using a sub-block as a minimum unit, this can improve reconstruction quality.

[0159] S206: Define a flag that corresponds to the region to be encoded for a first value.

[0160] For example, when it is determined to perform adaptive entropy coding skip processing on the region to be encoded in the feature map, the flag (for example, which may be represented by a skip flag, a flag, or a cube flag) that corresponds to the region to be encoded can be set to the first value; and then the S209 step is performed, i.e., the flag is encoded. In this way, the flag can be passed to the decoder, so that the decoder learns whether to perform adaptive entropy decoding skip processing on a corresponding region to be decoded. The first value can be set as required, for example, 1.

[0161] Based on the probability estimation result of the region to be encoded, after adaptive entropy coding jump processing is performed on the region to be encoded, a bit stream of the region to be encoded (i.e., a bit stream obtained through adaptive entropy coding jump processing) can be obtained.

[0162] It should be noted that, as shown in Figure 2, step S209 is performed after step S206, and step S205 is performed after step S209. It should be understood that, in the present invention, alternatively, step S205 may be performed first, then step S206 is performed, and then step S209 is performed. In other words, the present invention does not limit a flag encoding sequence and the performance of adaptive entropy coding skip processing in the region to be encoded. In the present invention, an example is used for description in which step S209 is performed after step S206, and step S205 is performed after step S209. Petition 870250082415, dated 12 / 09 / 2025, page 42 / 95 38 / 79

[0163] S207: Perform entropy coding processing on the region to be coded based on the probability estimation result of the region to be coded.

[0164] For example, when it is determined not to perform adaptive entropy coding jump processing in the region to be encoded in the feature map, entropy coding processing can be performed at each feature point in the region to be encoded based on the probability estimation result of each feature point in the region to be encoded, to obtain a bit stream (i.e., a bit stream obtained through entropy coding processing) of the region to be encoded.

[0165] S208: Define a flag that corresponds to the region to be encoded for a second value.

[0166] For example, when it is determined not to perform adaptive entropy coding skip processing in the region to be encoded in the feature map, the flag can be set to the second value; and then step S209 is performed, i.e., the flag is encoded. In this way, the flag can be passed to the decoder, so that the decoder learns whether to perform adaptive entropy decoding skip processing in a corresponding region to be decoded. The second value can be set as required, for example, 0.

[0167] It should be noted that the first value and the second value are different.

[0168] It should be noted that, as shown in Figure 2, step S209 is performed after step S208, and step S207 is performed after step S209. It should be understood that, in the present invention, alternatively, step S207 may be performed first, then step S208 is performed, and then step S209 is performed. In other words, the present invention does not limit a flag encoding sequence and the performance of entropy encoding processing in the region to be encoded. In the present invention, an example is used for description in which step S209 is performed after step S208, and step S207 is performed after step S209.

[0169] S209: Encode the flag.

[0170] In one possible mode, the flag can be encapsulated as a Petition 870250082415, dated 12 / 09 / 2025, page 43 / 95 39 / 79 syntax element in the bitstream generated through entropy encoding skip processing or in the bitstream generated through entropy encoding processing.

[0171] In one possible mode, the flag can be encoded to obtain a bitstream of the flag; and bitstreams of all regions to be encoded in the feature map and bitstreams of corresponding flags can form a bitstream of the image to be encoded.

[0172] It should be understood that there may be a plurality of flag encoding modes. This is not limited to the present invention. The details are described subsequently.

[0173] For example, steps S204 to S209 can be performed by the entropy encoding module.

[0174] For example, after all regions to be encoded in the feature map have been encoded according to steps S203 to S209, the bitstream of the image to be encoded can be obtained.

[0175] For example, the present invention provides a bitstream storage apparatus. The apparatus includes a receiver and at least one storage medium. The receiver is configured to receive a bitstream generated according to the encoding method in Figure 2; and at least one storage medium is configured to store the bitstream.

[0176] For example, the present invention provides a bitstream transmission apparatus. The apparatus includes a transmitter and at least one storage medium. The at least one storage medium is configured to store a bitstream generated according to the encoding method in Figure 2; and the transmitter is configured to obtain the bitstream from the storage medium and send the bitstream to a device-side device via a transmission medium.

[0177] For example, the present invention provides a bitstream delivery system. The system includes: at least one storage medium, configured to store at least one bitstream generated according to the encoding method in Figure 2; and a media streaming device, configured to obtain a target bitstream from at least one storage medium and send the target bitstream to a Petition 870250082415, dated 12 / 09 / 2025, page 44 / 95 40 / 79 device-side device, where the media streaming device includes a content server or a content delivery server.

[0178] Figure 3 is a diagram of an exemplary decoding process. Figure 3 shows the decoding process based on Figure 1a and Figure 1c. The decoding process in Figure 3 corresponds to the encoding process in Figure 2.

[0179] S301: Receive a bit stream.

[0180] For example, a feature map corresponding to the bit stream can be divided into a plurality of regions; then, each region is determined as a region to be decoded sequentially; and then, steps S302 to S306 are performed for the region to be decoded, to decode the region to be decoded. The following content uses a region to be decoded as an example for description. The region to be decoded can be a two-dimensional region, or it can be a three-dimensional region. The region to be decoded can include one or more feature points. It should be understood that the feature map is divided into the plurality of regions in a consistent manner in the decoding process and the encoding process.

[0181] S302: Determine a probability estimation result of the region to be decoded in the feature map that corresponds to the bitstream.

[0182] For example, the probability estimation module can perform probability estimation at each feature point in the feature map, and determine a probability estimation result for each feature point in the feature map, so that the probability estimation result for the region to be decoded in the feature map can be obtained. The probability estimation result for the region to be decoded can include probability estimation results for all feature points in the region to be decoded.

[0183] It should be understood that a process in step S302 corresponds to that in step S203, and the details are described subsequently.

[0184] S303: Decode the bit stream to obtain a flag that corresponds to the region to be decoded. Petition 870250082415, dated 12 / 09 / 2025, page 45 / 95 41 / 79

[0185] For example, the decoder can decode the bit stream to obtain the flag that corresponds to the region to be decoded.

[0186] It should be noted that, in one possible mode, step S302 can be performed first, and then step S303 is performed. In this case, an implementation of step S303 could be: decode the bitstream to obtain the flag that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded. In another possible mode, steps S302 and S303 can be performed in parallel. In this case, an implementation of step S303 could be: read the flag that corresponds to the region to be decoded from the bitstream. Specifically, a sequence for performing steps S302 and S303 is determined according to a flag encoding mode, and the details are described subsequently.

[0187] S304: When the flag is a first value, perform entropy decoding processing on at least one subblock in the region to be decoded or set a feature value of a feature point in at least one subblock to a specific value based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded.

[0188] For example, when the flag corresponding to the region to be decoded is the first value, it can be determined that the encoder performed adaptive entropy coding jump processing in the region to be decoded. Therefore, the decoder can perform adaptive entropy coding jump processing in the region to be decoded based on the probability estimation result of the region to be decoded, to obtain the feature value of the feature point in the region to be decoded.

[0189] For example, adaptive entropy decoding jump processing can be: perform entropy decoding processing on at least one sub-block in the region to be decoded or set the feature value of the feature point on at least one sub-block to the specific value based on the probability estimation result of the region to be decoded, to obtain the feature value of the feature point in the region to be decoded.

[0190] Specifically, the region to be decoded may include a Petition 870250082415, dated 12 / 09 / 2025, page 46 / 95 42 / 79 plurality of feature points, and at least one feature point can be grouped into a subblock. In this way, the region to be decoded can be divided into a plurality of subblocks, and each subblock can include at least one feature point. The probability estimation result of the region to be decoded includes probability estimation results of the plurality of subblocks, and the probability estimation result of each subblock can include probability estimation results of all feature points included in the subblock.For each sub-block, it is determined, based on the sub-block probability estimation result, whether to skip entropy decoding processing in the sub-block; and when it is determined not to skip entropy decoding processing in the sub-block, entropy decoding processing may be performed on all feature points in the sub-block based on the sub-block probability estimation result, to obtain feature values ​​for all feature points in the sub-block; or when it is determined to skip entropy decoding processing in the sub-block, entropy decoding processing on the feature points may be skipped, and feature values ​​for all feature points in the sub-block are set to a specific value.In this way, adaptive entropy decoding jump processing can be performed in the region to be decoded using a sub-block as a minimal unit, thereby improving decoding efficiency.

[0191] It should be understood that when the sub-block includes only one feature point, adaptive entropy decoding skip processing is performed using one feature point as a minimum unit. In this case, adaptive entropy decoding skip processing can be: perform entropy decoding processing on at least one feature point in the region to be decoded, or skip entropy decoding processing on at least one feature point based on the probability estimation result of the region to be decoded. In this way, the region to be decoded can be decoded using one feature point as a minimum unit. When compared to decoding the region to be decoded using a sub-block as a minimum unit, this can improve the quality of Petition 870250082415, dated 12 / 09 / 2025, page 47 / 95 43 / 79 reconstruction.

[0192] S305: When the flag is a second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded.

[0193] For example, when the flag corresponding to the region to be decoded is the second value, it can be determined that the encoder has performed entropy coding processing in the region to be decoded. Therefore, the decoder can perform entropy decoding processing at each feature point in the region to be decoded based on the probability estimation result of each feature point in the region to be decoded, to obtain a feature value for each feature point in the region to be decoded.

[0194] For example, steps S302 to S305 can be performed by the entropy decoding module.

[0195] S306: Perform reconstruction based on a feature value of a feature point in the feature map, to generate a reconstructed image.

[0196] In this way, after steps S302 to S305 are performed for all regions to be decoded in the feature map, feature values ​​of all feature points in the feature map can be obtained. Then, the feature dequantization module can dequantize the feature values ​​of all feature points in the feature map; and then, the feature decoding module can perform reconstruction based on dequantized feature values ​​of all feature points in the feature map, to obtain the reconstructed image.

[0197] For example, the feature decoding module can be implemented based on a neural network. For example, the feature decoding module can be implemented based on a convolutional neural network.

[0198] The following content describes encoding and decoding processes in detail. Petition 870250082415, dated 12 / 09 / 2025, page 48 / 95 44 / 79

[0199] Figure 4 is a diagram of an exemplary end-to-end image compression process.

[0200] As shown in Figure 4, for example, an encoding process may include feature extraction, quantization, extraction of auxiliary information, probability estimation, and arithmetic encoding. Feature extraction in Figure 4 can be performed by the feature extraction module in Figure 1c; quantization in Figure 4 can be performed by the feature quantization module in Figure 1c; probability estimation in Figure 4 can be performed by the encoder probability estimation module in Figure 1c; and arithmetic encoding in Figure 4 can be performed by the entropy encoding module in Figure 1c.

[0201] For example, the encoder may additionally include an auxiliary information extraction module, configured to perform auxiliary information extraction.

[0202] As shown in Figure 4, for example, a y-feature map is obtained after the feature extraction module performs feature extraction on an image to be encoded. A y-feature map is obtained after the y-feature map is quantized by the feature quantization module. Furthermore, after the auxiliary information extraction module performs auxiliary information extraction on the y-feature map, auxiliary information Z is obtained and emitted to the probability estimation module, and the probability estimation module performs probability estimation based on the auxiliary information Z, to obtain a probability estimation result of the y-feature map (i.e., a probability estimation result of the y-feature map).Therefore, the entropy encoding module can perform arithmetic encoding on the y-feature map based on the probability estimation result of the y-feature map (for details, see the descriptions in Figure 5a), to obtain a bitstream.

[0203] For example, the auxiliary information extraction module can alternatively perform auxiliary information extraction on the map of Petition 870250082415, dated 12 / 09 / 2025, page 49 / 95 45 / 79 characteristics y, to obtain auxiliary information Z. This is not limited to the present invention.

[0204] For example, after the auxiliary information Z is obtained, the auxiliary information Z can be additionally written to the bit stream, so that a probability estimation module of a decoder performs probability estimation.

[0205] As shown in Figure 4, for example, a decoding process may include arithmetic decoding, probability estimation, and image reconstruction. The arithmetic decoding in Figure 4 can be performed by the entropy decoding module in Figure 1c; the image reconstruction in Figure 4 can be performed by the feature decoding module in Figure 1c; and the probability estimation in Figure 4 can be performed by the decoder's probability estimation module in Figure 1c.

[0206] As shown in Figure 4, for example, the decoder can analyze the bit stream to obtain the auxiliary information Z, perform probability estimation based on the auxiliary information Z, and determine the probability estimation result of the feature map y. Then, arithmetic decoding can be performed based on the probability estimation result of the feature map y, to obtain the feature map y. Then, image reconstruction is performed based on the feature map y, to obtain a reconstructed image.

[0207] The following content describes in detail the encoding and decoding processes based on Figure 4.

[0208] Figure 5a is a diagram of an exemplary coding process. Figure 5a shows the coding process based on Figure 4.

[0209] S501: Get an image to be encoded.

[0210] S502: Extract a feature map from the image to be encoded.

[0211] For example, a feature map extracted by the feature extraction module for the image to be encoded can be referred to Petition 870250082415, dated 12 / 09 / 2025, page 50 / 95 46 / 79 as a y-feature map.

[0212] S503: Extract auxiliary information from the feature map.

[0213] In one possible mode, the y-feature map can be fed into the auxiliary information extraction module, and the auxiliary information extraction module extracts auxiliary information from each y[w][h][c] feature point in the y-feature map, to obtain auxiliary information Z, where w is an integer in the range of 1 to W, h is an integer in the range of 1 to H, and c is an integer in the range of 1 to C.

[0214] In one possible mode, the feature map y can be quantized first to obtain a feature map y ; and then the feature map y is fed into the auxiliary information extraction module, and the auxiliary information extraction module extracts auxiliary information from each feature point y [ w ][ h ][ c ] in the feature map y , to obtain auxiliary information Z.

[0215] Figure 5b is a diagram of an exemplary structure of an auxiliary information extraction module. As shown in Figure 5b, for example, the auxiliary information extraction module can be a neural network, and can include abs (absolute value function), a Conv (convolutional) layer, and a ReLU (activation) layer. Conv N*3*3 / 1 indicates that a convolutional kernel size of the convolutional layer is N*3*3, and a convolutional step is 1. Conv N*5*5 / 2 indicates that a convolutional kernel size of the convolutional layer is N*5*5, and a convolutional step is 2.

[0216] It should be understood that Figure 5b is only an example of the auxiliary information extraction module. The auxiliary information extraction module in the present invention may include more or fewer network layers / activation layers than those shown in Figure 5b. This is not limited in the present invention. Furthermore, the convolutional kernel size and the convolutional step of the convolutional layer are not limited in the present invention. In addition, the auxiliary information extraction module in the present invention may additionally include another network layer. This is also not limited in the present invention. Petition 870250082415, dated 12 / 09 / 2025, page 51 / 95 47 / 79

[0217] S504: Perform probability estimation based on auxiliary information, and determine a probability estimation result for a region to be coded on the feature map.

[0218] For example, the probability estimation module can perform probability estimation at each feature point y[w][h][c] in the feature map y based on the auxiliary information of each feature point y[w][h][c] in the feature map y, to obtain a probability estimation result for each feature point y[w][h][c] in the feature map y, and additionally obtain the probability estimation result for the region to be encoded.

[0219] In one possible mode, the probability estimation result for each feature point can be a probability P.

[0220] In one possible mode, when the probability estimation module performs Gaussian distribution estimation, the probability estimation result may include a Gaussian distribution parameter: a mean μ and a variance σ.

[0221] Figure 5c is a diagram of an exemplary structure of a probability estimation module. As shown in Figure 5c, for example, the probability estimation module can be a neural network, and can include a Conv (convolutional) layer and a ReLU (activation) layer. Conv 2Mx3x3 / 1 indicates that a convolutional kernel size of the convolutional layer is 2Mx3x3, and a convolutional step is 1. Conv Nx5x5 / 2 indicates that a convolutional kernel size of the convolutional layer is Nx5x5, and a convolutional step is 2.

[0222] For example, the feature map y can be divided into a plurality of cubes (cube) using wlxhlxcl as a basic unit (w1 is an integer in the range of 1 to W, h1 is an integer in the range of 1 to H, and c1 is an integer in the range of 1 to C). When c1=1, a cube means a block.

[0223] The following content uses an example for description in which a region to be encoded is a cube. For each cube, it is determined, based on a probability estimation result of the cube, whether it realizes Petition 870250082415, dated 12 / 09 / 2025, page 52 / 95 48 / 79 Adaptive entropy encoding jump processing on the cube. See the following descriptions of steps S505 and S506.

[0224] S505: Determine a reconstructed feature value of a feature point in the region to be encoded after adaptive entropy coding jump processing is performed on the region to be encoded.

[0225] In one possible mode, for a feature point y[w][h][c] in the cube, it is determined whether a probability P of a feature point value y[w][h][c] being an original feature value (which can be represented by k1) is greater than or equal to T0; and when the probability P of the feature point value y[w][h][c] being k1 is greater than or equal to T0, a specific value (which can be represented by k2) is used as a reconstructed feature value of the feature point y[w][h][c]; or when the probability P of the feature point value y[w][h][c] being k1 is less than T0, the original feature value of the feature point y[w][h][c] can be used as a reconstructed feature value of the feature point y[w][h][c].

[0226] In one possible mode, when probability estimation is Gaussian distribution estimation, and the result of probability estimation is a Gaussian distribution parameter: a mean μ and a variance σ, for a feature point y[w][h][c] in the cube, it can be determined whether a Gaussian distribution parameter of the feature point y[w][h][c] satisfies the following condition: abs(μ-k1) + σ < T1; and when the Gaussian distribution parameter of the feature point y[w][h][c] satisfies abs(μ-k1) + σ < T1, k2 can be used as a reconstructed feature value of the feature point y[w][h][c]; or when the Gaussian distribution parameter of the characteristic point y[w][h][c] does not satisfy abs(μ-k 1) + σ < T1, an original characteristic value of y[w][h][c] Petition 870250082415, dated 12 / 09 / 2025, page 53 / 95 49 / 79 can be used as a reconstructed feature value of the feature point y[w][h][c].

[0227] For example, T0 and T1 are predefined thresholds, and T0 and T1 can each be any value between 0 and 1. This is not limited in the present invention. For example, T0 is 0.99, 0.98, 0.97, or 0.95. For example, T1 is 0.2, 0.3, or 0.4. Furthermore, k2 can be any value, and is specifically defined as required, for example, 0, 1, 2, or 3. This is not limited in the present invention.

[0228] In this way, a reconstructed feature value for each feature point in the cube can be obtained.

[0229] S506: Determine, based on a reconstructed feature value and an original feature value of the region to be encoded, whether to perform adaptive entropy coding skip processing in the region to be encoded.

[0230] Next, an error between a reconstructed feature value and an original feature value of a cube can be calculated. Then, based on the error, it is determined whether to perform adaptive entropy coding jump processing on the cube. For example, a minimum mean square error (MES) between a reconstructed feature value and an original feature value of a feature point on the cube can be calculated as an error.

[0231] For example, when the error between the reconstructed feature value and the original feature value of the cube is greater than Thr (a predefined error threshold), it can be determined that performing adaptive entropy coding skip processing on the region to be encoded introduces an artifact. In this case, adaptive entropy coding skip processing is not performed on the cube, and steps S510, S511, and S509 can be performed. When the error between the reconstructed feature value and the original feature value of the cube is less than or equal to Thr, it is determined that performing adaptive entropy coding skip processing on the region to be encoded does not introduce an artifact. In this case, adaptive entropy coding skip processing can be performed on the cube, and steps S508, S511, and S507 (including S507a and S507b) Petition 870250082415, dated 12 / 09 / 2025, page 54 / 95 50 / 79 can be performed.

[0232] The original feature value of the cube may include an original feature value of a feature point included in the cube, and the reconstructed feature value of the cube may include a reconstructed feature value of a feature point included in the cube.

[0233] S507: Perform entropy coding processing on at least one subblock in the region to be coded or skip entropy coding processing on at least one subblock based on the probability estimation result of the region to be coded.

[0234] Step S507 may include steps S507a and S507b.

[0235] For example, the region to be encoded can be divided into a plurality of sub-blocks; and then step S507a or S507b is performed for a first sub-block in the plurality of sub-blocks included in the region to be encoded (each sub-block in the region to be encoded can be determined as the first sub-block in sequence):

[0236] S507a: When a probability estimation result from the first subblock satisfies a predefined condition, skip entropy encoding processing on all feature points in the first subblock.

[0237] S507b: When a probability estimation result from the first subblock does not satisfy a predefined condition, perform entropy encoding processing on all feature points in the first subblock based on probability estimation results from all feature points in the first subblock.

[0238] For example, when the probability estimation result of the first sub-block satisfies the predefined condition, while entropy encoding processing on all feature points in the first sub-block is skipped, feature values ​​of all feature points in the first sub-block can be additionally set to a specific k2 value, to facilitate subsequent image reconstruction by the encoder.

[0239] In one possible mode, the predefined condition may be that a probability P of a sub-block is greater than or equal to T2.

[0240] When the sub-block includes a plurality of points Petition 870250082415, dated 12 / 09 / 2025, page 55 / 95 51 / 79 characteristic, the probability P of the sub-block can be an average of probabilities P of all characteristic points in the sub-block, or it can be a maximum or minimum value of probabilities P of all characteristic points in the sub-block. This is not limited in the present invention. When the sub-block includes only one characteristic point, the probability P of the sub-block is a probability P of the one characteristic point included in the sub-block.

[0241] T2 can be a predefined threshold. T2 can be any value between 0 and 1. For example, T2 is 0.99, 0.98, 0.97, or 0.95. This is not limited in the present invention. In one possible mode, T2 can be equal to T0. In another possible mode, T2 can not be equal to T0.

[0242] In one possible mode, when the probability estimation result is a Gaussian distribution parameter, the predefined condition may be that a Gaussian distribution parameter of a subblock satisfies abs (μ-k 1) + σ < T3 .

[0243] For example, when the sub-block includes a plurality of feature points, a sub-block mean μ can be a mean of means μ of all feature points in the sub-block, and a sub-block variance σ can be a mean of variances σ of all feature points in the sub-block. In one possible mode, a sub-block mean μ can be a maximum or a minimum value of means μ of all feature points in the sub-block, and a sub-block variance σ can be a maximum or a minimum value of variances σ of all feature points in the sub-block. For example, when the sub-block includes only one feature point, a mean μ of the sub-block can be a mean μ of the feature point included in the sub-block, and a variance σ of the sub-block can be a variance σ of the feature point included in the sub-block.

[0244] T3 can be a predefined threshold. T3 can be any value between 0 and 1. For example, T3 is 0.2, 0.3, or 0.4. This is not limited in the present invention. In one possible mode, T3 can be equal to T1. In another possible mode, T3 can not be equal to T1.

[0245] In one possible mode, the encoder and decoder can pre-agree on a predefined condition to be used and a threshold (T2 or T3) used. Petition 870250082415, dated 12 / 09 / 2025, page 56 / 95 52 / 79 in the predefined condition. In this way, the encoder does not need to write T2 or T3 in a bit stream.

[0246] In one possible mode, the encoder can pre-construct a list of threshold candidates (which may include T2 and T3) and synchronize the list of threshold candidates with the decoder. Each threshold in the list of threshold candidates corresponds to an index number, the length of the list of threshold candidates can be defined as N, and N is an integer greater than or equal to 1. One mode of constructing the list of threshold candidates is not limited in the present invention. The encoder can write an index number of a threshold in the list of candidates in a bitstream for transmission to the decoder. Specifically, the index number can be stored in a sequence header, an image header, a slice header, or supplemental enhancement information (SEI) and transmitted to the decoder.It should be understood that the encoder can alternatively write an index number in a bit stream using another method. This is not limited to the present invention.

[0247] In one possible mode, the encoder can define a threshold (which may include T2 or T3) based on image complexity. In this way, the encoder can write the threshold (T2 or T3) used in the predefined condition in a bitstream. Specifically, the threshold is written in the bitstream, for example, the threshold can be stored in a sequence header, an image header, a slice header, or SEI and transmitted to the decoder. It should be understood that the encoder can alternatively write an index number in a bitstream using another method. This is not limited to the present invention.

[0248] S508: Define a flag that corresponds to the region to be encoded for a first value.

[0249] S509: Perform entropy coding processing on the region to be coded based on the probability estimation result of the region to be coded.

[0250] S510: Define a flag that corresponds to the region to be encoded for a second value.

[0251] S511: Encode the flag.

[0252] For example, one possible way to encode the flag could be: Petition 870250082415, dated 12 / 09 / 2025, page 57 / 95 53 / 79 write the flag to the bitstream after flag binarization. It should be understood that when both the first and second values ​​are binary values, the flag can be written directly to the bitstream.

[0253] For example, one possible flag encoding method could be: determine a flag value probability based on the probability estimation result of the region to be encoded; and then encode the flag based on the flag value probability. The flag value probability is a probability that the flag is the first value or a probability that the flag is the second value.

[0254] For example, one possible way to determine the probability of the flag value based on the probability estimation result of the region to be encoded could be: determine the probability of the flag value based on a sum of probability estimation results of a plurality of sub-blocks in the region to be encoded.

[0255] In one possible mode, when the probability estimation result is a Gaussian distribution parameter, Gaussian distribution parameters of all feature points in all subblocks in the region to be encoded can be accumulated to obtain a Gaussian distribution parameter of the flag. A flag mean is pf = ^μ, and a flag variance is Gf =^σ. Then, a probability of a Gaussian distribution with mean Pf and variance Gf in an interval of [0.5, 0.5] is used as the probability that the flag is the first value.

[0256] In one possible mode, a sum of probabilities P of all feature points in all sub-blocks in the region to be encoded is calculated; and then an average probability is calculated based on the sum, and is used as the probability that the flag is the first value.

[0257] For example, one possible way to determine the flag value probability based on the probability estimation result of the region to be encoded could be: determine the flag value probability based on a number of feature points where entropy encoding processing is skipped in the region to be encoded and a total number of feature points in the region to be encoded, where the number of points Petition 870250082415, dated 12 / 09 / 2025, page 58 / 95 54 / 79 of feature points in which entropy encoding processing is skipped in the region to be encoded is determined based on probability estimation results of a plurality of feature points included in the region to be encoded.

[0258] In one possible mode, the number of feature points at which entropy encoding processing is skipped in the region to be encoded can be determined in a process of performing step S505, that is, when a probability P of a feature point value í- [ W][ h ][c ] being k1 is greater than or equal to T0, the feature point y[w][h][c] is determined as a feature point at which entropy encoding processing is skipped.

[0259] In one possible mode, the number of feature points at which entropy encoding processing is skipped in the region to be encoded can be determined in a process carrying out step S505, that is, when a Gaussian distribution parameter of a feature point y[w][h][c] satisfies abs(μ-k1) + σ < T1, the feature point y[w][h][c] is determined as a feature point at which entropy encoding processing is skipped.

[0260] In one possible mode, the number of feature points at which entropy encoding processing is skipped in the region to be encoded can be determined in a process of performing step S507, i.e., when a probability estimation result of a subblock satisfies the predefined condition, a feature point y[w][h][c] in the subblock is determined as a feature point at which entropy encoding processing is skipped.

[0261] For example, a ratio between the number of feature points at which entropy encoding processing is skipped in the region to be encoded and the total number of feature points in the region to be encoded can be calculated, and the ratio is used as the probability that the flag is the first value.

[0262] For example, after steps S505 to S511 are completed Petition 870250082415, dated 12 / 09 / 2025, page 59 / 95 55 / 79 for each region to be encoded, it can be determined whether the region to be encoded is a last region in the feature map. If so, the procedure is terminated, and a bitstream of the image to be encoded is obtained. If not, it returns to step S505.

[0263] For example, after steps S505, S506, S507a, S508, and S510 are completed for each region to be encoded, it can be determined whether the region to be encoded is a last region in the feature map. If so, steps S507b, S509, and S511 are performed on all regions to be encoded to obtain a bitstream of the image to be encoded. If not, it returns to step S505. In other words, for all regions to be encoded, flag encoding processing and entropy encoding are finally performed. In this case, a flag encoding mode could be the following:

[0264] For example, a possible implementation of flag encoding in the S511 step for all regions to be encoded could be: after flag binarization, group and encode flags that correspond to all regions to be encoded, wherein flags of k regions to be encoded (namely, k cubes) are grouped into one group for encoding; and write the flags to the bitstream.

[0265] An example in which k is 8 is used for description. When eight flags in a group are each the first value, a group flag is set to a third value (for example, the third value is 0), and the third value is written to the bitstream; otherwise, a group flag is set to a fourth value (the fourth value is different from the third value, for example, the fourth value is 1), the fourth value is written to the bitstream, and binary values ​​of the eight flags in the group are written to the bitstream. A corresponding syntax table is shown in Table 1 after the flags corresponding to all regions to be encoded have been encoded. Table 1 is the syntax table of the flags corresponding to all regions to be encoded.

[0266] As shown in Table 1, number_of_cube_flag_group represents a number of groups after all regions are grouped; group_flag represents a group flag; cube_flag represents a flag that corresponds to a region to be encoded (namely, the flags in steps S206, S208 and S209, in steps S303, S304 and S305, and in steps S508, S510 and S511); Petition 870250082415, dated 12 / 09 / 2025, page 60 / 95 56 / 79 number_of_elements_in_group represents the number of flags in a last group; and uf(1) indicates that a matching symbol is written in a bit stream. Table 1 Syntax Description for( i = 0; i < number_of_cube_flag_group-1; i++ ) { group_flag[ i ] uf(1) if(group_flag[ i ] ) for( j = 0; j < 8; j++ ) cube_flag[ i ][ j ] uf(1)} group_flag [number_of_cube_flag_group-1 ] uf(1) if(group_flag[number_of_cube_flag_group-1 ] ) for( j = 0; j < number_of_elements_in_group; j++ ) cube_flag[ i ][ j ] uf(1)}

[0267] For example, a possible implementation of flag encoding in the S511 step for all regions to be encoded could be: writing flags to the bitstream after flag binarization. It should be understood that when the first and second values ​​are binary values, the flag can be written directly to the bitstream.

[0268] For example, a possible implementation of flag encoding in step S511 for all regions to be encoded could be: determine flag value probabilities based on probability estimation results of the regions to be encoded; and then encode the flags based on the flag value probabilities. The flag value probability is a probability that the flag is the first value or a probability that the flag is the second value.

[0269] Figure 6 is a diagram of an exemplary decoding process. Figure 6 shows the decoding process based on Figure 4. The decoding process in Figure 6 corresponds to the encoding process in Figure 5a.

[0270] S601: Receive a bit stream. Petition 870250082415, dated 12 / 09 / 2025, page 61 / 95 57 / 79

[0271] S602: Decode the bitstream to obtain auxiliary information from a feature map, and determine a probability estimation result of a region to be decoded in the feature map based on the auxiliary information from the feature map.

[0272] For example, the encoder writes auxiliary information Z of a feature map y into a bitstream, so that the decoder can perform entropy decoding processing on the bitstream and decode the bitstream to obtain the auxiliary information Z of the feature map y, then perform probability estimation based on the auxiliary information z of the feature map y, to determine a probability estimation result of the feature map, and additionally obtain the probability estimation result of a region to be decoded in the feature map. This process is similar to that of the encoder in step S504, and the details are not described again in the present invention.

[0273] S603: Decode the bit stream to obtain a flag that corresponds to the region to be decoded.

[0274] For example, if the encoder writes the flag to the bitstream after flag binarization, the decoder can directly read the flag that corresponds to the region to be decoded from the bitstream.

[0275] For example, if the encoder groups and encodes flags that correspond to all regions to be encoded and writes the flags to the bitstream after flag binarization, the decoder can perform decoding based on the syntax table of the flags that correspond to all regions to be encoded shown in Table 1, to read the flag that corresponds to the region to be decoded. First, a group flag value is read; and, when the group flag value is a third value, it can be determined that all k flags in a group are a first value, i.e., flags of k regions to be decoded (namely, k cubes) are the first value; or when the group flag value is a fourth value, k flags in a group are read from the bitstream through decoding. In the present invention, k is the same as that on one side of the encoder. For example, k in the encoder is 8, and k in the decoder is also 8. Petition 870250082415, dated 12 / 09 / 2025, page 62 / 95 58 / 79

[0276] For example, if the encoder encodes the flag based on a flag value probability, the decoder can first determine the flag value probability based on the probability estimation result of the region to be decoded that is determined in step S602 and then decode the bit stream to obtain the flag that corresponds to the region to be decoded based on the flag value probability.

[0277] For example, the encoder and decoder can pre-agree on a way to determine the probability of the flag value.

[0278] If the encoder determines the probability of the flag value based on a sum of probability estimation results from a plurality of feature points in the region to be encoded, the decoder can determine the probability of the flag value based on a sum of probability estimation results from a plurality of feature points in the region to be decoded. For details, see the descriptions of step S511 above. The details are not described again in the present invention.

[0279] If the encoder determines the flag value probability based on a number of feature points at which entropy encoding processing is skipped in the region to be encoded and a total number of feature points in the region to be encoded, the decoder can determine, based on the probability estimation result of the region to be decoded, a number of feature points at which entropy decoding processing is skipped in the region to be decoded; and determine the flag value probability based on the number of feature points at which entropy decoding processing is skipped in the region to be decoded and a total number of feature points in the region to be decoded.

[0280] For example, the encoder and decoder may pre-agree on a way to determine a feature point at which entropy decoding processing is skipped (or entropy encoding processing is skipped) in the region to be decoded (or in the region to be encoded). When the encoder determines a feature point y[w][h][c] whose probability P of a value being k1 is greater than or equal to T0 Petition 870250082415, dated 12 / 09 / 2025, page 63 / 95 59 / 79 as a feature point where entropy encoding processing is skipped, the decoder can determine the feature point y[w][h][c] whose probability P of the value being k1 is greater than or equal to T0 as a feature point where entropy decoding processing is skipped. When the encoder determines a feature point y[w][h][c] whose Gaussian distribution parameter satisfies abs(μ-k1) + σ < T1 as a feature point where entropy encoding processing is skipped, the decoder can determine the feature point y[w][h][c] whose Gaussian distribution parameter satisfies abs(μ-k1) + σ < T1 as a feature point where entropy decoding processing is skipped.When the encoder determines a feature point y[w][h][c] included in a subblock whose probability estimation result satisfies a predefined condition as a feature point where entropy encoding processing is skipped, the decoder can determine the feature point y[w][h][c] included in the subblock whose probability estimation result satisfies the predefined condition as a feature point where entropy decoding processing is skipped.

[0281] S604: When the flag is the first value, perform entropy decoding processing on at least one subblock in the region to be decoded, or set a feature value of a feature point in at least one subblock to a specific value based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded.

[0282] Step S604 may include steps S604a and S604b.

[0283] For example, if the flag is the first value, the decoder can perform step S604a or S604b for a first sub-block in a plurality of sub-blocks included in the region to be decoded:

[0284] S604a: When a probability estimation result from the first sub-block satisfies the predefined condition, skip processing of Petition 870250082415, dated 12 / 09 / 2025, page 64 / 95 60 / 79 entropy decoding at all feature points in the first subblock, and setting feature values ​​of all feature points in the first subblock to a specific k2 value.

[0285] S604b: When a probability estimation result from the first subblock does not satisfy the predefined condition, perform entropy decoding processing on all feature points in the first subblock based on probability estimation results from all feature points in the first subblock.

[0286] The decoder can determine, based on T2 or T3 read from the bitstream, whether the probability estimation result of the first subblock satisfies the predefined condition. Alternatively, the decoder can search a list of threshold candidates for T2 or T3 based on an index number read from the bitstream and then determine whether the probability estimation result of the first subblock satisfies the predefined condition. Certainly, the decoder can alternatively determine, based on T2 or T3 agreed upon in advance, whether the probability estimation result of the first subblock satisfies the predefined condition.

[0287] S605: When the flag is a second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded.

[0288] S606: Perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

[0289] For example, for steps S605 and S606, refer to the preceding descriptions of steps S305 and S306. The details are not described again in the present invention.

[0290] For example, the feature decoding module can be implemented based on a neural network of any structure, for example, a fully connected network, a convolutional neural network, or a cyclic neural network. Optionally, the neural network can use a multi-layered deep neural network structure to achieve better Petition 870250082415, dated 12 / 09 / 2025, pages 65 / 95 61 / 79 estimation effect.

[0291] In this way, a transmission flag is encoded at a cube level, indicating whether to perform entropy encoding and decoding jump processing based on a probability estimation result for a current cube, thus avoiding a case where an artifact is generated by executing a jump algorithm based on a probability estimation result.

[0292] In one possible mode, in an encoding process, if a probability estimation result is a Gaussian distribution parameter, after a feature map y and a probability estimation result of the feature map y are obtained, a predictor of each feature point y[w][h][c] in the feature map y is defined for a corresponding mean μ. In this way, a feature map v can be obtained. Then, a residual between feature values ​​of corresponding feature points in the feature map y and the feature map v can be calculated, to obtain a feature map r. Then, the feature map r is quantized to obtain a feature map r. Subsequently, entropy encoding processing can be performed on the feature map f with reference to steps S505 to S511.Correspondingly, in a decoding process, after steps S604 and S605 are performed, a feature value of a feature point in the feature map r is obtained. Then, a predictor of each feature point [w][h][c] in the feature map f is set to a corresponding mean μ. In this way, the feature map v can be obtained. Then, feature values ​​of corresponding feature points in the feature map re in the feature map v can be added to obtain a feature map ye, then step S606 is performed.

[0293] It should be noted that whether a feature map used for entropy encoding processing is the y-feature map or the r-feature map is not limited in the present invention.

[0294] Figure 7 is a diagram of a compression process of Petition 870250082415, dated 12 / 09 / 2025, page 66 / 95 62 / 79 exemplified end-to-end image. Figure 7 shows the end-to-end image compression process based on Figure 4. The probability estimation in Figure 7 is a Gaussian distribution estimation. In Figure 7, context information can be used to optimize a Gaussian distribution parameter μ.

[0295] As shown in Figure 7, for example, a coding process may include feature extraction, quantization, extraction of ancillary information, probability estimation, first predictor estimation, second predictor estimation, extraction of context information, and arithmetic coding.

[0296] For example, the encoder may additionally include a context module (configured to perform context information extraction in the encoding process in Figure 7), a predictor estimation module (configured to perform first predictor estimation in the encoding process in Figure 7), and a prediction module (configured to perform second predictor estimation in the encoding process in Figure 7). The probability estimation module and the predictor estimation module may be combined into one. This is not limited to the present invention.

[0297] As shown in Figure 7, for example, a y-feature map is obtained after the feature extraction module performs feature extraction on an image to be encoded. Then, after the auxiliary information extraction module performs auxiliary information extraction on the y-feature map, auxiliary information z is obtained (auxiliary information z can be obtained by performing auxiliary information extraction on the y-feature map by the auxiliary information extraction module, and then the auxiliary information z can be quantized to obtain the auxiliary information z), and is output to the probability estimation module.The probability estimation module performs Gaussian distribution estimation based on the auxiliary information z, to obtain a Gaussian distribution parameter σ [ w ][ h ][ c ] of each feature point y [ w ][ h ][ c ] in the feature map y. Additionally, the predictor estimation module can perform first predictor estimation on the auxiliary information z, and output a feature map F to the module. Petition 870250082415, dated 12 / 09 / 2025, page 67 / 95 63 / 79 prediction. Then, the prediction module can perform a second predictor estimation on a feature G obtained by performing context information extraction based on the feature map F and the context module, and output a Gaussian distribution parameter μ[w][h][c] for each feature point y[w][h][c]. For a feature point, the context module performs context information extraction based on a reconstructed feature value from a spatially encoded feature point adjacent to the feature point.

[0298] For example, a predictor of each feature point y[ w ][ h ][ c ] in the feature map y can be defined for a corresponding mean / / [ w ][ h ][ c ]. In this way, a feature map v can be obtained. Then, a residual between feature values ​​of corresponding feature points in the feature map y and in the feature map v can be calculated, to obtain a feature map r. Then, the feature map r is quantized to obtain a feature map f. Furthermore, in a reconstruction process, feature values ​​of corresponding feature points in the feature map f and in the feature map v can be added to obtain a feature map y. A feature value of a coded feature point y [ w ][ h ][ c ] in the feature map y is a reconstructed feature value of the coded feature point.Clearly, a Gaussian distribution parameter of the feature map y is (μ,σ). The feature map r satisfies a Gaussian distribution with a mean of 0 and a variance of σ, that is, the Gaussian distribution parameter of the feature map r is (0,σ).

[0299] So, the entropy encoding module can perform arithmetic encoding on the feature map f based on a probability estimation result of the feature map f (for details, see the descriptions in Figure 5a), to obtain a bit stream 1; and perform arithmetic encoding on the auxiliary information z, to obtain a bit stream 2. Both bit stream 1 and bit stream 2 are part of a stream of Petition 870250082415, dated 12 / 09 / 2025, page 68 / 95 64 / 79 bits of the image to be encoded.

[0300] As shown in Figure 7, for example, a decoding process may include arithmetic decoding, probability estimation, first predictor estimation, second predictor estimation, context information extraction, and image reconstruction.

[0301] For example, the decoder may additionally include a context module (configured to perform context information extraction in the decoding process in Figure 7), a predictor estimation module (configured to perform first predictor estimation in the decoding process in Figure 7), and a prediction module (configured to perform second predictor estimation in the decoding process in Figure 7). The probability estimation module and the predictor estimation module may be combined into one. This is not limited to the present invention.

[0302] As shown in Figure 7, for example, the decoder can perform arithmetic decoding on bit stream 2, analyze bit stream 2 to obtain auxiliary information Z, and perform Gaussian distribution estimation based on the auxiliary information Z, to obtain a Gaussian distribution parameter σ[w][h][c] for each feature point r[w][h][c] in the feature map r. Furthermore, the predictor estimation module can perform first predictor estimation on the auxiliary information Z, and output the feature map F to the prediction module. Then, the prediction module can perform second predictor estimation on the feature G obtained by performing context information extraction based on the feature map F and the context module, and output a Gaussian distribution parameter μ[w][h][c] for each feature point y[w][h][c].For a feature point, the context module performs context information extraction based on a feature value reconstructed from a spatially decoded feature point adjacent to the feature point.

[0303] For example, arithmetic decoding can be performed on bit stream 1 based on the Gaussian distribution parameter σ [ w][h ][ c ]of each feature point r[ w ][ h][c ]in the feature map r, for Petition 870250082415, dated 12 / 09 / 2025, page 69 / 95 65 / 79 analyze bit stream 1 to obtain the feature map r. Then, a predictor of each feature point r [w][h][c] in the feature map r can be set to a corresponding mean / / [w][h][c]. In this way, the feature map v can be obtained. Next, feature values ​​of corresponding feature points in the feature map r in the feature map v can be added to obtain the feature map y. A feature value of a decoded feature point y[w][h][c] in the feature map y is a reconstructed feature value of the decoded feature point.

[0304] For an entropy encoding process and an entropy decoding process for the r-feature map, see the descriptions in the embodiments in Figure 5a and Figure 6. The details are not described again in the present invention.

[0305] It should be understood that the context module is an optional module. When the encoder (decoder) does not include the context module, the encoder (decoder) performs first predictor estimation on the auxiliary information z using the predictor estimation module, and outputs the Gaussian distribution parameter μ\w][h][c] for each characteristic point y[w][h][c].

[0306] It should be understood that the auxiliary information extraction module is an optional module. When the encoder (decoder) does not include the auxiliary information extraction module, the encoder (decoder) can invoke the context module to perform context information extraction, and then the probability estimation module performs probability estimation based on the context information to determine a probability estimation result.

[0307] Figure 8 is a diagram of an exemplary encoding process. Figure 8 shows the encoding process based on Figure 1a and Figure 1c. In the embodiment in Figure 8, an example of encoding an image is used for description. The image can be a standalone image, or it can be a frame in a video. This is not limited to Petition 870250082415, dated 12 / 09 / 2025, pp. 70 / 95 66 / 79 present invention.

[0308] S801: Get an image to be encoded.

[0309] S802: Extract a feature map from the image to be encoded.

[0310] S803: Determine a probability estimation result for a region to be coded on the feature map.

[0311] For example, for steps S801 to S803, see the preceding descriptions of steps S201 to S203.

[0312] For example, the probability estimation result of the region to be coded on the feature map can be determined based on context information and / or auxiliary information.

[0313] For example, a y-feature map or an r-feature map can be divided into a plurality of cubes (cube) using wlxhl xc1 as a basic unit (w1 is an integer in the range of 1 to W, h1 is an integer in the range of 1 to H, and c1 is an integer in the range of 1 to C). When c1=1, a cube means a block.

[0314] The following content uses an example for description in which a region to be encoded is a cube.

[0315] S804: Determine, based on an original feature value of the region to be encoded and a reconstructed feature value of the region to be encoded that is determined after entropy encoding processing in the region to be encoded is skipped, whether to skip entropy encoding processing in the region to be encoded.

[0316] For example, the reconstructed feature value of the region to be encoded can first be determined whether entropy encoding processing in the region to be encoded is skipped; and then the reconstructed feature value of the region to be encoded is compared to the original feature value to determine whether entropy encoding processing in the region to be encoded is skipped.

[0317] For example, by comparing the reconstructed feature value with the original feature value of the region to be encoded, it can be determined whether an artifact is introduced after entropy encoding processing in the region to be encoded is skipped. When it is Petition 870250082415, dated 12 / 09 / 2025, page 71 / 95 67 / 79. If it is determined, based on a comparison result between the reconstructed feature value and the original feature value of the region to be encoded, that an artifact is introduced after entropy encoding processing in the region to be encoded is skipped, entropy encoding processing in the region to be encoded may not be skipped. In this case, steps S808, S809, S810, and S807 may be performed. If it is determined, based on a comparison result between the reconstructed feature value and the original feature value of the region to be encoded, that no artifact is introduced after entropy encoding processing in the region to be encoded is skipped, entropy encoding processing in the region to be encoded may be skipped. In this case, steps S806, S809, S810, and S805 may be performed.This can prevent the introduction of an artifact when the encoder adaptively determines entropy coding skip processing.

[0318] The original feature value of the region to be encoded may include original feature values ​​from a plurality of feature points included in the region to be encoded, and the reconstructed feature value of the region to be encoded may include reconstructed feature values ​​from the plurality of feature points included in the region to be encoded.

[0319] Specifically, reconstructed feature values ​​of all feature points y[w][h][c] in a cube can be set to a specific value k2. Then, an error between a reconstructed feature value and an original feature value of the cube can be calculated. Then, it is determined, based on the error, whether to skip entropy encoding processing in the cube. For example, a minimum mean square error (MES) between a reconstructed feature value and an original feature value of a feature point in the cube can be calculated as an error.

[0320] For example, when the error between the reconstructed feature value and the original feature value of the cube is greater than Thr (a predefined error threshold), it can be determined that entropy encoding processing on the cube does not need to be skipped, and step S807 can Petition 870250082415, dated 12 / 09 / 2025, page 72 / 95 68 / 79 to be performed; or when the error between the reconstructed feature value and the original feature value of the cube is less than or equal to Thr, it is determined that entropy encoding processing on the cube can be skipped, and step S805 can be performed.

[0321] S805: Skip entropy encoding processing in the region to be encoded.

[0322] For example, when entropy encoding processing at all feature points in the region to be encoded is skipped, feature values ​​of all feature points in the region to be encoded can be set to the specific value k2, to facilitate subsequent image reconstruction by the encoder.

[0323] S806: Define a flag that corresponds to the region to be encoded for a first value.

[0324] It should be noted that, as shown in Figure 8, steps S809 and S810 are performed after step S806, and step S805 is performed after steps S809 and S810. It should be understood that, in the present invention, alternatively, step S805 can be performed first, then step S806 is performed, and then steps S809 and S810 are performed. In other words, the present invention does not limit a sequence of encoding the flag and skipping entropy encoding processing in the region to be encoded.

[0325] S807: Perform entropy coding processing on the region to be coded based on the probability estimation result of the region to be coded.

[0326] S808: Define a flag that corresponds to the region to be encoded for a second value.

[0327] For example, for steps S805 to S808, refer to the descriptions in the preceding embodiment. The details are not described again in the present invention.

[0328] It should be noted that, as shown in Figure 8, steps S809 and S810 are performed after step S808, and step S807 is performed after steps S809 and S810. It should be understood that, in the present invention, alternatively, step S807 may be performed first, then step S808 is performed, and then steps S809 and S810 are performed. In other words, the present invention does not limit a flag encoding and implementation sequence. Petition 870250082415, dated 12 / 09 / 2025, page 73 / 95 69 / 79 of entropy encoding processing in the region to be encoded.

[0329] S809: Determine a probability of the flag value that corresponds to the region to be encoded based on the probability estimation result of the region to be encoded, wherein the probability of the flag value is either a probability that the flag is the first value or a probability that the flag is the second value.

[0330] For example, in a possible mode of the S809 step, the flag value probability can be determined based on a sum of probability estimation results from a plurality of feature points in the region to be encoded. For details, see the descriptions in the preceding embodiment. The details are not described again in the present invention.

[0331] For example, in a possible mode of step S809, a number of feature points at which entropy encoding processing is skipped in the region to be encoded can be determined based on the probability estimation result of the region to be encoded; and the flag value probability is determined based on the number of feature points at which entropy encoding processing is skipped in the region to be encoded and a total number of feature points in the region to be encoded. For details, see the descriptions in the preceding embodiment. The details are not described again in the present invention.

[0332] S810: Encode the flag based on the probability of the flag value.

[0333] For example, for step S810, see the preceding descriptions. The details are not described again in the present invention.

[0334] Figure 9 is a diagram of an exemplary decoding process. Figure 9 shows the decoding process based on Figure 1a and Figure 1c. The decoding process in Figure 9 corresponds to the encoding process in Figure 8.

[0335] S901: Get a bitstream.

[0336] S902: Determine a probability estimation result of a region to be decoded in a feature map that corresponds to the bitstream.

[0337] S903: Determine a probability value of a flag that Petition 870250082415, dated 12 / 09 / 2025, pp. 74 / 95 70 / 79 corresponds to the region to be decoded based on the probability estimation result of the region to be decoded, where the flag value probability is either the probability that the flag is a first value or the probability that the flag is a second value.

[0338] For example, in an implementation of step S903, a number of feature points at which entropy decoding processing is skipped in the region to be decoded can be determined based on the probability estimation result of the region to be decoded; and the flag value probability is determined based on the number of feature points at which entropy decoding processing is skipped in the region to be decoded and a total number of feature points in the region to be decoded. For details, see the preceding descriptions. The details are not described again in the present invention.

[0339] For example, in an implementation of step S903, the flag value probability can be determined based on a sum of probability estimation results from a plurality of feature points in the region to be decoded. For details, see the preceding descriptions. The details are not described again in the present invention.

[0340] S904: Decode the bit stream to obtain the flag that corresponds to the region to be decoded based on the flag value probability.

[0341] S905: When the flag is the first value, set a feature value of a feature point in the region to be decoded to a specific value.

[0342] S906: When the flag is the second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded.

[0343] After steps S905 and S906 have been performed for all regions to be decoded in the feature map, if feature values ​​of all feature points in a map of Petition 870250082415, dated 12 / 09 / 2025, pp. 75 / 95 If 71 / 79 feature values ​​y are obtained, step S907 can be performed directly; if feature values ​​of all feature points in a feature map r are obtained, predictors of all feature points in the feature map r can be set to a mean μ, to obtain a feature map v; and then the feature values ​​of all feature points in the feature map r and the feature values ​​of all feature points in the feature map v are added, to obtain a feature map y, and then step S907 is performed.

[0344] S907: Perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

[0345] Figure 10 is a diagram of an exemplary machine vision task based on encoding and decoding. Figure 10 is shown based on Figure 1b.

[0346] For an entropy encoding and decoding process in Figure 10, see the descriptions in the preceding embodiment. After a feature map y is obtained through entropy decoding, the machine vision task is executed based on the feature map y, to implement classification, target recognition, semantic segmentation, and the like.

[0347] For example, the present invention provides an encoder. The encoder may include: a capture module, configured to obtain an image to be encoded; a feature extraction module, configured to extract a feature map from the image to be encoded; a probability estimation module, configured to determine a probability estimation result for a region to be encoded in the feature map; and an entropy encoding module, configured to: determine, based on an original feature value of the region to be encoded and on Petition 870250082415, dated 12 / 09 / 2025, pp. 76 / 95 72 / 79 a reconstructed feature value that is determined after adaptive entropy coding skip processing is performed on the region to be coded, if adaptive entropy coding skip processing is performed on the region to be coded; if so, perform entropy coding processing on at least one sub-block in the region to be coded or skip entropy coding processing on at least one sub-block based on the probability estimation result of the region to be coded, and set a flag that corresponds to the region to be coded for a first value; or if not, perform entropy coding processing on the region to be coded based on the probability estimation result of the region to be coded, and set a flag that corresponds to the region to be coded for a second value; and encode the flag.

[0348] It should be understood that the encoder in the present invention can additionally perform the steps in the encoding processes in Figure 2 and Figure 5a.

[0349] For example, the present invention provides a decoder. The decoder may include: a receiver module, configured to receive a bit stream; An entropy decoding module, configured to: determine a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream, and decode the bit stream to obtain a flag that corresponds to the region to be decoded; and when the flag is a first value, perform entropy decoding processing on at least one sub-block in the region to be decoded or set a feature value of a feature point in at least one sub-block to a specific value based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded;or when the flag is a second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value from a feature point in the region to be decoded; and a feature decoding module, configured to perform reconstruction based on a feature value from a feature point; Petition 870250082415, dated 12 / 09 / 2025, page 77 / 95 73 / 79 feature on the feature map, to obtain a reconstructed image.

[0350] It should be understood that the decoder in the present invention can additionally perform the steps in the decoding processes in Figure 3 and Figure 6.

[0351] For example, the present invention provides an encoder. The encoder may include: a capture module, configured to obtain an image to be encoded; a feature extraction module, configured to extract a feature map from the image to be encoded; a probability estimation module, configured to determine a probability estimation result for a region to be encoded in the feature map; and an entropy encoding module, configured to: determine, based on an original feature value of the region to be encoded and a reconstructed feature value that is determined after entropy encoding processing on the region to be encoded is skipped, whether to skip entropy encoding processing on the region to be encoded; if so, skip entropy encoding processing on the region to be encoded, and set a flag that corresponds to the region to be encoded for a first value; or if not, perform entropy encoding processing on the region to be encoded based on the probability estimation result of the region to be encoded, and set a flag that corresponds to the region to be encoded for a second value;Determine a flag value probability based on the probability estimation result of the region to be encoded, where the flag value probability is either the probability that the flag is the first value or the probability that the flag is the second value; and encode the flag based on the flag value probability.

[0352] It should be understood that the encoder in the present invention can additionally perform the steps in the encoding process in Figure 8.

[0353] For example, the present invention provides a decoder. The decoder may include: a receiver module, configured to receive a bit stream; Petition 870250082415, dated 12 / 09 / 2025, pp. 78 / 95 74 / 79 an entropy decoding module, configured to: determine a probability estimation result of a region to be decoded in a feature map that corresponds to the bitstream; determine a flag value probability that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded, where the flag value probability is either a probability that the flag is a first value or a probability that the flag is a second value; decode the bitstream to obtain the flag based on the flag value probability; and when the flag is the first value, set a feature value of a feature point in the region to be decoded to a specific value;or when the flag is the second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and a feature decoding module, configured to perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

[0354] It should be understood that the decoder in the present invention can additionally perform the steps in the decoding process in Figure 9.

[0355] In one example, Figure 11 is a block diagram of an apparatus 1100 according to an embodiment of the present invention. The apparatus 1100 may include a processor 1101 and a transceiver / transceiver pin 1102. Optionally, the apparatus 1100 additionally includes a memory 1103.

[0356] The components of the 1100 device are coupled together via a 1104 bus. In addition to a data bus, the 1104 bus also includes a power bus, a control bus, and a status signal bus. However, for clear description, several buses are referred to as the 1104 bus in the figure.

[0357] Optionally, memory 1103 can be configured to store instructions in the modes of the preceding method. The Petition 870250082415, dated 12 / 09 / 2025, page 79 / 95 The 75 / 79 processor 1101 can be configured to: execute instructions in memory 1103, control a receiver pin to receive a signal, and control a transmit pin to send a signal.

[0358] Device 1100 can be the electronic device or a chip of the electronic device in the embodiments of the preceding method.

[0359] For example, the electronic device can be a terminal device, or it can be a server.

[0360] All related content of the steps, in the embodiments of the preceding method, can be cited in function descriptions of the corresponding functional modules. The details are not described again in the present invention.

[0361] One embodiment of the present invention further provides a chip, including one or more interface circuits and one or more processors. The one or more processors receive or send data through the one or more interface circuits. When the one or more processors execute computer instructions, an electronic device is enabled to perform the steps of the preceding related method to implement the encoding and decoding methods in the preceding embodiments. The interface circuit is a transceiver / transceiver pin 1102.

[0362] One embodiment additionally provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions. When the computer instructions are run on an electronic device, the electronic device is enabled to perform the antecedent related method steps to implement the encoding and decoding methods in the antecedent embodiments.

[0363] An embodiment additionally provides a computer program product. The computer program product includes computer instructions. When the computer instructions are executed by a computer or processor, the computer is enabled to perform the related antecedent steps to implement the encoding and decoding methods in the antecedent embodiments.

[0364] In addition, one embodiment of the present invention further provides a device. Specifically, the device may be a chip, Petition 870250082415, dated 12 / 09 / 2025, pages 80 / 95 76 / 79 a component or a module. The device may include a processor and a memory that are connected to each other. The memory is configured to store computer executable instructions. When the device is run, the processor can execute the computer executable instructions stored in the memory, to enable the chip to perform the encoding and decoding methods in the modes of the preceding method.

[0365] The electronic device, computer-readable storage media, computer program product, or chip, provided in embodiments, is configured to perform the corresponding method provided above. Therefore, for beneficial effect that can be achieved, refer to the beneficial effect in the corresponding method provided above. The details are not described in the present invention.

[0366] Based on the descriptions of the preceding implementations, a person skilled in the art may understand that, for the purpose of convenient and brief description, the division into the preceding functional modules is used as an example for illustration. During actual application, the preceding functions may be allocated to different functional modules and implemented as required, i.e., an internal structure of a device is divided into different functional modules to implement all or some of the functions described above.

[0367] In various embodiments provided in the present invention, it should be understood that the apparatus and method disclosed can be implemented in other ways. For example, the embodiment of the apparatus described is only one example. For example, the division into modules or units is only a logical function division and may be another division during actual implementation. For example, a plurality of units or components may be combined or integrated into another apparatus, or some features may be ignored or not implemented. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed may be implemented through some interfaces. Indirect couplings or communication connections between the apparatuses or units may be implemented in electronic, mechanical or other forms.

[0368] The units described as separate parts may or may not be physically separate, and parts displayed as units may be one or Petition 870250082415, dated 12 / 09 / 2025, pp. 81 / 95 77 / 79 more physical units, that is, they can be located in one place, or they can be distributed across different locations. Some or all of the units can be selected based on actual requirements to achieve the objectives of the solutions of the modalities.

[0369] Furthermore, the functional units, in embodiments of the present invention, may be integrated into a processing unit, or each of the units may exist physically on its own, or two or more units may be integrated into one unit. The integrated unit may be implemented in a hardware form, or it may be implemented in the form of a software functional unit.

[0370] Any content in embodiments of the present invention and any content in the same embodiment may be freely combined. Any combination of the foregoing content is within the scope of the present invention.

[0371] When the integrated unit is implemented in the form of a functional software unit and marketed or used as a standalone product, the integrated unit can be stored on a readable storage medium. Based on such understanding, the technical solutions of embodiments of the present invention essentially, or the part that contributes to conventional technology, or all or some of the technical solutions can be implemented in the form of a software product. The software product is stored on a storage medium and includes various instructions to instruct a device (which may be a single-chip microcomputer, a chip, or similar) or a processor to perform all or some of the steps of the methods described in embodiments of the present invention.The preceding storage media includes any media that can store program code, such as a USB flash drive, a removable hard disk, read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0372] The preceding content describes embodiments of the present invention with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely examples and not limitations. Inspired by Petition 870250082415, dated 12 / 09 / 2025, pages 82 / 95 78 / 79 In the present invention, a person skilled in the art may additionally make modifications without departing from the purposes of the present invention and the scope of protection of the claims, and all modifications must be included in the protection of the present invention.

[0373] The algorithm methods or steps, described in combination with the content disclosed in embodiments of the present invention, may be implemented by hardware, or may be implemented by a processor executing software instructions. The software instruction may include a corresponding software module. The software module may be stored in random access memory (RAM), flash memory, read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable read-only programmable read-only memory (EEPROM), a register, a hard disk, a removable hard disk, a compact disc read-only memory (CD-ROM), or any other form of storage media well known in the art.For example, a storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Certainly, the storage medium can be a component of the processor. The processor and the storage medium can be arranged in an ASIC.

[0374] A person skilled in the art should be aware that, in one or more of the preceding examples, the functions described in embodiments of the present invention may be implemented by hardware, software, firmware, or any combination thereof. When the functions are implemented by software, the functions may be stored on a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. The computer-readable medium includes a computer-readable storage medium and a communication medium. The communication medium includes any media that enables a computer program to be transmitted from one location to another. The storage medium may be any available media accessible by a general-purpose or special-purpose computer. Petition 870250082415, dated 12 / 09 / 2025, pp. 83 / 95 79 / 79

[0375] The preceding content describes embodiments of the present invention with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely examples and not limitations. Inspired by the present invention, a person skilled in the art may further make modifications without departing from the purposes of the present invention and the scope of protection of the claims, and all modifications must be included in the protection of the present invention. Petition 870250082415, dated 12 / 09 / 2025, pages 84 / 95

Claims

1 / 8 CLAIMS 1. Coding method, CHARACTERIZED in that the method comprises: obtaining an image to be coded; extracting a feature map from the image to be coded, and determining a probability estimation result of a region to be coded in the feature map; determining, based on an original feature value of the region to be coded and a reconstructed feature value of the region to be coded that is determined after adaptive entropy coding skip processing is performed on the region to be coded, whether adaptive entropy coding skip processing should be performed on the region to be coded;If yes, perform entropy encoding processing on at least one sub-block in the region to be encoded, or skip entropy encoding processing on at least one sub-block based on the probability estimation result of the region to be encoded, and set a flag that corresponds to the region to be encoded for a first value; or if no, perform entropy encoding processing on the region to be encoded based on the probability estimation result of the region to be encoded, and set a flag that corresponds to the region to be encoded for a second value; and encode the flag.

2. Method, according to claim 1, CHARACTERIZED in that encoding the flag comprises: determining a flag value probability based on the probability estimation result of the region to be encoded, wherein the flag value probability is a probability that the flag is the first value or a probability that the flag is the second value; and encoding the flag based on the flag value probability.

3. Method, according to claim 1 or 2, CHARACTERIZED in that determining, based on the original feature value of the region to be encoded and the reconstructed feature value of the region to be encoded, which is determined after adaptive entropy coding jump processing is performed on the region to be encoded, whether adaptive entropy coding jump processing should be performed on the region to be encoded comprises: determining an error between the reconstructed feature value and the original feature value; and when the error is greater than an error threshold, determining whether adaptive entropy coding jump processing should be performed on the region to be encoded; or when the error is less than or equal to an error threshold, determining whether adaptive entropy coding jump processing should not be performed on the region to be encoded.

4. A method according to claim 2, characterized in that the region to be encoded comprises a plurality of subblocks, and the probability estimation result of the region to be encoded comprises probability estimation results of the plurality of subblocks; and determining the flag value probability based on the probability estimation result of the region to be encoded comprises: determining the flag value probability based on a sum of the probability estimation results of the plurality of subblocks in the region to be encoded; for a first subblock in the plurality of subblocks: when a probability estimation result of the first subblock satisfies a predefined condition, skipping entropy encoding processing at all feature points in the first subblock;Or, when a probability estimation result from the first subblock does not satisfy a predefined condition, perform entropy encoding processing on all feature points in the first subblock based on probability estimation results from all feature points in the first subblock.

5. Decoding method, CHARACTERIZED by the fact that the method comprises: receiving a bit stream; determining a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream, and decoding the bit stream to obtain a flag that corresponds to the region to be decoded; when the flag is a first value, performing entropy decoding processing on at least one sub-block in the region to be decoded or setting a feature value of a feature point in at least one sub-block to a specific value based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded;Or, when the flag is a second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

6. Method, according to claim 5, CHARACTERIZED in that decoding the bit stream to obtain the flag that corresponds to the region to be decoded comprises: determining a probability of the flag value based on the probability estimation result of the region to be decoded, wherein the probability of the flag value is a probability that the flag is the first value or a probability that the flag is the second value; and decoding the bit stream to obtain the flag that corresponds to the region to be decoded based on the probability of the flag value.

7. Method, according to claim 5, CHARACTERIZED in that the region to be decoded comprises a plurality of subblocks, and the probability estimation result of the region to be decoded comprises probability estimation results of the plurality of subblocks; and determining the flag value probability based on the probability estimation result of the region to be decoded comprises: determining the flag value probability based on a sum of the probability estimation results of the plurality of subblocks in the region to be decoded. Petition 870250082415, dated 12 / 09 / 2025, pp. 87 / 95 4 / 8 8. A method according to claim 5, characterized in that the region to be decoded comprises a plurality of subblocks, the subblock comprises at least one feature point, and the probability estimation result of the region to be decoded comprises probability estimation results of a plurality of feature points; and determining the flag value probability based on the probability estimation result of the region to be decoded comprises: determining, based on the probability estimation results of the plurality of feature points comprised in the region to be decoded, a number of feature points at which entropy decoding processing is skipped in the region to be decoded;and determine the probability of the flag value based on the number of feature points at which entropy decoding processing is skipped in the region to be decoded and a total number of feature points in the region to be decoded.

9. Encoding method, CHARACTERIZED by the fact that the method comprises: obtaining an image to be encoded; extracting a feature map from the image to be encoded, and determining a probability estimation result of a region to be encoded in the feature map; determining, based on an original feature value of the region to be encoded and a reconstructed feature value of the region to be encoded that is determined after entropy coding processing in the region to be encoded is skipped, whether to skip entropy coding processing in the region to be encoded; if so, skipping entropy coding processing in the region to be encoded, and setting a flag that corresponds to the region to be encoded for a first value;or, if not, perform entropy coding processing on the region to be coded based on the probability estimation result of the region to be coded, and define a flag that corresponds to the region to be coded for a second value; Petition 870250082415, dated 12 / 09 / 2025, page 88 / 95 5 / 8 determine a flag value probability based on the probability estimation result of the region to be coded, where the flag value probability is a probability that the flag is the first value or a probability that the flag is the second value; and encode the flag based on the flag value probability.

10. Decoding method, CHARACTERIZED by the fact that the method comprises: receiving a bit stream; determining a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream; determining a probability value of a flag that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded, wherein the probability value of the flag is a probability that the flag is a first value or a probability that the flag is a second value; decoding the bit stream to obtain the flag based on the probability value of the flag; when the flag is the first value, setting a feature value of a feature point in the region to be decoded to a specific value;Or, when the flag is the second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

11. Encoder, CHARACTERIZED in that the encoder comprises: an acquisition module, configured to obtain an image to be encoded; a feature extraction module, configured to extract a feature map from the image to be encoded; Petition 870250082415, dated 12 / 09 / 2025, page 89 / 95 6 / 8 a probability estimation module, configured to determine a probability estimation result of a region to be encoded in the feature map; and an entropy encoding module, configured to: determine, based on an original feature value of the region to be encoded and a reconstructed feature value of the region to be encoded that is determined after adaptive entropy coding skip processing is performed on the region to be encoded, whether adaptive entropy coding skip processing is performed on the region to be encoded;If yes, perform entropy encoding processing on at least one sub-block in the region to be encoded, or skip entropy encoding processing on at least one sub-block based on the probability estimation result of the region to be encoded, and set a flag that corresponds to the region to be encoded for a first value; or if no, perform entropy encoding processing on the region to be encoded based on the probability estimation result of the region to be encoded, and set a flag that corresponds to the region to be encoded for a second value; and encode the flag.

12. Decoder, CHARACTERIZED in that the decoder comprises: a receiver module, configured to receive a bit stream; an entropy decoding module, configured to: determine a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream, and decode the bit stream to obtain a flag that corresponds to the region to be decoded; and when the flag is a first value, perform entropy decoding processing on at least one sub-block in the region to be decoded or set a feature value of a feature point in at least one sub-block to a specific value based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded;or when the flag is a second value, perform entropy decoding processing in the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and Petition 870250082415, dated 12 / 09 / 2025, pp. 90 / 95 7 / 8 a feature decoding module, configured to perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

13. Encoder, CHARACTERIZED in that the encoder comprises: an acquisition module, configured to obtain an image to be encoded; a feature extraction module, configured to extract a feature map from the image to be encoded; a probability estimation module, configured to determine a probability estimation result of a region to be encoded in the feature map; and an entropy encoding module, configured to: determine, based on an original feature value of the region to be encoded and a reconstructed feature value of the region to be encoded that is determined after entropy encoding processing in the region to be encoded is skipped, whether to skip entropy encoding processing in the region to be encoded;If yes, skip entropy encoding processing in the region to be encoded, and set a flag that corresponds to the region to be encoded for a first value; or if no, perform entropy encoding processing in the region to be encoded based on the probability estimation result of the region to be encoded, and set a flag that corresponds to the region to be encoded for a second value; determine a flag value probability based on the probability estimation result of the region to be encoded, where the flag value probability is either a probability that the flag is the first value or a probability that the flag is the second value; and encode the flag based on the flag value probability.

14. Decoder, CHARACTERIZED in that the decoder comprises: a receiver module, configured to receive a bit stream; an entropy decoding module, configured to: determine a probability estimation result of a region to be decoded in a feature map that corresponds to the bit stream; Petition 870250082415, dated 12 / 09 / 2025, page 91 / 95 8 / 8 determine a probability value of a flag that corresponds to the region to be decoded based on the probability estimation result of the region to be decoded, wherein the probability value of the flag is a probability that the flag is a first value or a probability that the flag is a second value; decode the bit stream to obtain the flag based on the probability value of the flag; and when the flag is the first value, set a feature value of a feature point in the region to be decoded to a specific value;or when the flag is the second value, perform entropy decoding processing on the region to be decoded based on the probability estimation result of the region to be decoded, to obtain a feature value of a feature point in the region to be decoded; and a feature decoding module, configured to perform reconstruction based on a feature value of a feature point in the feature map, to obtain a reconstructed image.

15. Bitstream storage apparatus, CHARACTERIZED in that the apparatus comprises a receiver and at least one storage medium; the receiver is configured to receive a bitstream; and at least one storage medium is configured to store the bitstream, wherein the bitstream is generated according to the encoding method as defined in any one of claims 1 to 4, or is generated according to the encoding method as defined in claim 9. Petition 870250082415, dated 12 / 09 / 2025, pp. 92 / 95

Images