An image compression system and an image encoding method, decoding method, compression method

Abstract

The present disclosure provides an image compression system and an image encoding method, decoding method and compression method, wherein the image compression system comprises an encoding network, a decoding network and an entropy model; the encoding network comprises a main encoder and a bypass encoder, and the encoding network is used for encoding an image to be compressed, and the bypass encoder adopts an orthogonal linear constraint; the decoding network comprises a main decoder and a bypass decoder, and the main decoder and the bypass decoder are connected in parallel, and the decoding network is used for decoding a compressed image binary code stream, and the bypass decoder adopts an orthogonal linear constraint; and the entropy model is used for predicting a mean value and a variance of a Gaussian distribution. Through the present disclosure, the bypass encoder and the bypass decoder are introduced into the image compression system, the training time of an end-to-end image compression system is efficiently reduced, the image compression code rate overhead is reduced, and the quality of a reconstructed image is improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of image processing, in particular, to an image compression system and an image encoding method, a decoding method and a compression method. BACKGROUND

[0002] In recent years, with the rapid development of deep learning technology, end-to-end image compression models based on deep learning have attracted extensive attention and rapid progress. Such models can provide more adaptive quantization units for multi-dimensional quantizers by constructing complex nonlinear transformation structures, breaking through the construction limitations of traditional transform coding methods, and theoretically having potential advantages in image compression performance. However, existing deep learning-based image compression methods still face some significant challenges in practical applications, of which the most prominent pain point is the excessively long training time. Existing end-to-end image compression models usually require a large amount of data and a high-dimensional feature learning process, which results in an extremely long training period for the model. In particular, when training on large-scale datasets, the complexity and computational demand of deep neural networks are further increased, resulting in a very significant increase in training time. This long training process not only brings great pressure on resource utilization and computational cost, but also seriously affects the development cycle and application promotion of the model. Therefore, how to effectively shorten the training time of the model while ensuring the image compression effect has become a key technical problem that needs to be solved in the field of image compression.

[0003] To solve this problem, recent researches optimize the structure of the encoder and decoder of the image compression model, such as Dailan Hen et al. in 2022 Proceedings of the IEEE / CVF Conference on Computer Vision and Pattern Recognition, “ELIC: Efficient Learned Image Compression with Unevenly Grouped Space-Channel Contextual Adaptive Coding”, which replaces the normalization layer with stacked residual blocks at the encoding and decoding end to achieve efficient training and higher rate-distortion performance. However, this method still requires an iterative training of two million steps to converge to a better performance result. SUMMARY

[0004] In view of the defects in the prior art, the purpose of the present disclosure is to provide an image compression system and an image encoding method, a decoding method and a compression method.

[0005] To achieve the above object, according to a first aspect of the present disclosure, an image compression system is provided, comprising an encoding network, a decoding network and an entropy model;

[0006] The encoding network comprises a main encoder and a bypass encoder, the encoding network is used for encoding a to-be-compressed image, and the bypass encoder adopts orthogonal linear constraint;

[0007] The decoding network comprises a main decoder and a bypass decoder, the main decoder and the bypass decoder are connected in parallel, the decoding network is used for decoding a compressed image binary code stream, and the bypass decoder adopts orthogonal linear constraint;

[0008] The entropy model is used for predicting a mean value and a variance of a Gaussian distribution.

[0009] Optionally, the main encoder adopts an end-to-end image compression encoding network, comprising a plurality of convolution blocks or a plurality of Transformer blocks.

[0010] Optionally, the bypass encoder comprises a plurality of orthogonal linear wavelet encoding units, each of the orthogonal linear wavelet encoding units comprises a cascaded discrete wavelet forward transform and an orthogonal linear layer.

[0011] Optionally, the main decoder adopts a decoding network of an end-to-end image compression system, and the main decoder comprises a plurality of convolution blocks or a plurality of Transformer blocks.

[0012] Optionally, the bypass decoder comprises a plurality of orthogonal linear wavelet decoding units, each of the orthogonal linear wavelet decoding units comprises a cascaded discrete wavelet inverse transform and an orthogonal linear layer.

[0013] Optionally, a training loss function of the encoding network, the decoding network and the entropy model is as follows:

[0014]

[0015] Wherein, represents distortion, x represents the to-be-compressed image, represents a reconstructed image, λ1 represents a first balance parameter, λ2 represents a second balance parameter, and R1 represents code rate, represents a loss term for constraining orthogonality of the orthogonal linear layer;

[0016]

[0017] Wherein, represents a feature of a kth element, represents a predicted distribution of the kth element, θ k ={φ k ,ψk} represents a set of learnable parameters, represents a prediction distribution of side information;

[0018]

[0019] where I represents a matrix with diagonal values of 1, W enc,j represents a coefficient of a quadrature linear layer of a quadrature linear wavelet encoding unit of the bypass encoder, W enc,j represents a coefficient of a quadrature linear layer of a quadrature linear wavelet decoding unit of the bypass decoder.

[0020] According to a second aspect of the present disclosure, an image encoding method is provided, comprising:

[0021] inputting a to-be-compressed image into an encoding network to determine a first feature map of the to-be-compressed image, the first feature map representing a three-dimensional feature map;

[0022] quantizing each feature value in the first feature map into an integer to determine a second feature map;

[0023] inputting the second feature map into an entropy model, and modeling a distribution of the second feature map as a Gaussian distribution to determine a Gaussian distribution of the second feature map;

[0024] determining a predicted mean and variance of the Gaussian distribution according to the Gaussian distribution of the second feature map,

[0025] determining a probability estimate of the second feature map according to the predicted mean and variance of the Gaussian distribution;

[0026] arithmetic coding the probability estimate of the second feature map to determine a compressed image binary code stream.

[0027] Optionally, the inputting a to-be-compressed image into an encoding network to determine a first feature map of the to-be-compressed image comprises:

[0028] inputting the to-be-compressed image into a main encoder to determine a main feature map;

[0029] inputting the to-be-compressed image into a bypass encoder to determine a bypass feature map;

[0030] combining the main feature map and the bypass feature map to determine the first feature map of the to-be-compressed image.

[0031] Optionally, the inputting a to-be-compressed image into a bypass encoder to determine a first bypass feature map comprises:

[0032] inputting the compressed image into a first orthogonal linear wavelet encoding unit, and sequentially taking the output of each previous orthogonal linear wavelet encoding unit as the input of a next orthogonal linear wavelet encoding unit, and taking the output of a last orthogonal linear wavelet encoding unit as the first bypass feature;

[0033] wherein the input of each orthogonal linear wavelet encoding unit is taken as the input of a discrete wavelet forward transform of each orthogonal linear wavelet encoding unit, the output of the discrete wavelet forward transform is taken as the input of an orthogonal linear layer of each orthogonal linear wavelet encoding unit, and the output of the orthogonal linear layer is taken as the output of each orthogonal linear wavelet encoding unit.

[0034] According to a third aspect of the present disclosure, a method for image decoding is provided, which is used for decoding a compressed image binary code stream obtained by the method for image encoding according to the second aspect of the present disclosure, and includes:

[0035] obtaining a compressed image binary code stream;

[0036] inputting the compressed image binary code stream into an entropy model to determine the mean and variance of a Gaussian distribution of a predicted second feature map;

[0037] performing arithmetic decoding according to the mean and variance of the Gaussian distribution of the predicted second feature map to determine the second feature map;

[0038] inputting the second feature map into a decoding network to determine a reconstructed image.

[0039] Optionally, the inputting the second feature map into the decoding network to determine the reconstructed image includes:

[0040] inputting the second feature map into a main decoder to determine a main decoded image;

[0041] inputting the second feature map into a bypass decoder to determine a bypass decoded image;

[0042] combining the main decoded image and the bypass decoded image to determine the reconstructed image.

[0043] Optionally, the inputting the second feature map into the bypass decoder to determine the bypass decoded image includes:

[0044] inputting the second feature map into a first orthogonal linear wavelet decoding unit, and sequentially taking the output of each previous orthogonal linear wavelet decoding unit as the input of a next orthogonal linear wavelet decoding unit, and taking the output of a last orthogonal linear wavelet decoding unit as the bypass decoded image;

[0045] The input quantity of each orthogonal linear wavelet decoding unit is used as the input quantity of the inverse discrete wavelet transform of each orthogonal linear wavelet decoding unit, the inverse discrete wavelet transform outputs a discrete wavelet feature, and the output quantity of the inverse discrete wavelet transform is used as the input quantity of the orthogonal linear layer of each orthogonal linear wavelet decoding unit, the orthogonal linear layer outputs a second transform feature, and the output quantity of the orthogonal linear layer is used as the output quantity of each orthogonal linear wavelet decoding unit.

[0046] According to a fourth aspect of the present disclosure, an image compression method is provided, comprising:

[0047] An image to be compressed is obtained.

[0048] The image to be compressed is encoded based on any one of the image encoding methods provided in the second aspect of the present disclosure, and a compressed image binary code stream is obtained.

[0049] The compressed image binary code stream is decoded based on any one of the image decoding methods provided in the third aspect of the present disclosure, and a reconstructed image is obtained.

[0050] According to a fifth aspect of the present disclosure, a terminal is provided, comprising a memory and a processor, the memory stores a program executed by the processor, and the processor can execute any one of the following methods when executing the program:

[0051] The image encoding method provided in any one of the second aspect of the present disclosure, or

[0052] The image decoding method provided in any one of the third aspect of the present disclosure, or

[0053] The image compression method provided in the fourth aspect of the present disclosure.

[0054] According to a sixth aspect of the present disclosure, a non-transitory computer readable storage medium is provided, which stores a computer program, and the program is executed by a processor, and the program can execute the image encoding method provided in the second aspect of the present disclosure, or the image decoding method provided in the third aspect of the present disclosure, or the image compression method provided in the fourth aspect of the present disclosure.

[0055] Compared with the prior art, the embodiments of the present disclosure have at least one of the following beneficial effects:

[0056] Through the above technical solution, the bypass encoder is introduced into the encoding network of the image compression system, and the orthogonal linear constraint is adopted, the bypass decoder is introduced into the decoding network of the image compression system, and the orthogonal linear constraint is adopted, which can efficiently reduce the training time of the end-to-end image compression system, and can save the code rate under the condition of obtaining the same reconstructed image quality.

[0057] Embodiments of the present disclosure introduce multiple orthogonal linear wavelet coding units in the bypass decoder, which can efficiently reduce the training time of the end-to-end image compression system. BRIEF DESCRIPTION OF DRAWINGS

[0058] Other features, objects, and advantages of the present disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the attached drawings:

[0059] Figure 1 is a structural schematic diagram of an image compression system according to an exemplary embodiment.

[0060] Figure 2 is a flowchart of an image encoding method according to an exemplary embodiment.

[0061] Figure 3 is a flowchart of an image decoding method according to an exemplary embodiment.

[0062] Figure 4 is a flowchart of an image compression method according to an exemplary embodiment.

[0063] Figure 5 is a block diagram of an image compression system according to an exemplary embodiment.

[0064] Figure 6 is a flowchart of an image compression system training method according to an exemplary embodiment. DETAILED DESCRIPTION

[0065] The present disclosure will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art further understand the present disclosure, but do not limit the present disclosure in any form. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present disclosure. These all belong to the protection scope of the present disclosure.

[0066] Figure 1 is a structural schematic diagram of an image compression system according to an exemplary embodiment.

[0067] As shown in Figure 1 , the present disclosure provides an image compression system, including an encoding network, a decoding network and an entropy model.

[0068] The encoding network includes a main encoder and a bypass encoder, and the encoding network is used for encoding the image to be compressed. The bypass encoder adopts orthogonal linear constraint.

[0069] The bypass encoder can accelerate the training and convergence of the encoding network and further accelerate the convergence of the encoding network and the entropy model based on the orthogonal linear constraint.

[0070] The decoding network includes a main decoder and a bypass decoder connected in parallel, the decoding network is used for decoding a compressed image binary code stream, and the bypass decoder adopts the orthogonal linear constraint.

[0071] The bypass decoder can accelerate the training and convergence of the decoding network and further accelerate the convergence of the decoding network and the entropy model based on the orthogonal linear constraint.

[0072] The entropy model is used to predict the mean and variance of the Gaussian distribution.

[0073] In the present disclosure, the main encoder and the main decoder can adopt the encoding network and the decoding network in the existing end-to-end image compression method, including but not limited to a convolutional network, a residual network, and a Transformer network; and the upper model can adopt any entropy model, such as a hyper-prior entropy model, a spatial auto-regressive entropy model, and a channel auto-regressive entropy model.

[0074] Through the above technical solutions, the bypass encoder is introduced into the encoding network of the image compression system and the orthogonal linear constraint is adopted, the bypass decoder is introduced into the decoding network of the image compression system and the orthogonal linear constraint is adopted, the training time of the end-to-end image compression system can be efficiently reduced, and the code rate can be saved under the condition of obtaining the same reconstructed image quality.

[0075] In a possible embodiment, the main encoder adopts the encoding network of the end-to-end image compression, including multiple convolutional blocks or multiple Transformer blocks.

[0076] The main encoder can adopt any existing encoding network of the end-to-end image compression method, including but not limited to a convolutional network, a residual network, and a Transformer network.

[0077] In the present disclosure, the main encoder is composed of 4 convolutional layers with a downsampling ratio of 2 and 3 ReLU nonlinear activation units connected alternately, specifically, a convolutional layer with upsampling, a ReLU nonlinear activation unit, a convolutional layer with upsampling, a ReLU nonlinear activation unit, a convolutional layer with upsampling, a ReLU nonlinear activation unit, and a convolutional layer with upsampling are connected in sequence.

[0078] The bypass encoder includes multiple orthogonal linear wavelet encoding units, each orthogonal linear wavelet encoding unit includes a discrete wavelet forward transform and an orthogonal linear layer connected in sequence.

[0079] wherein the coefficients of the orthogonal linear layer of each orthogonal linear wavelet encoding unit are W enc,j .

[0080] In the present disclosure, the bypass encoder is cascaded by four orthogonal linear wavelet encoding units.

[0081] In a possible embodiment, the main decoder adopts a decoding network of an end-to-end image compression system, and the main decoder comprises a plurality of convolutional blocks or a plurality of Transformer blocks.

[0082] In the present disclosure, the main decoder is cascaded by four convolutional layers with an upsampling factor of 2 and three ReLU nonlinear activation units alternately, specifically, a convolutional layer with upsampling, a ReLU nonlinear activation unit, a convolutional layer with upsampling, a ReLU nonlinear activation unit, a convolutional layer with upsampling, a ReLU nonlinear activation unit, and a convolutional layer with upsampling are sequentially cascaded.

[0083] The bypass decoder comprises a plurality of orthogonal linear wavelet decoding units, each of which comprises a cascaded discrete wavelet inverse transform and an orthogonal linear layer.

[0084] wherein the coefficients of the orthogonal linear layer of each orthogonal linear wavelet decoding unit are W dec,j .

[0085] In the present disclosure, the bypass decoder is sequentially cascaded by four orthogonal linear wavelet decoding units.

[0086] By introducing a plurality of orthogonal linear wavelet encoding units in the bypass decoder, the training time of the end-to-end image compression system can be efficiently reduced.

[0087] The image compression system provided by the present disclosure can be used as an image compression model, and further adopts a random gradient descent method for stage optimization.

[0088] In a possible embodiment, the training loss function of the encoding network, the decoding network, and the entropy model is:

[0089]

[0090] wherein, denotes distortion, x denotes a to-be-compressed image, denotes a reconstructed image, λ1 denotes a first balance parameter, λ2 denotes a second balance parameter, R1 denotes code rate, denotes a loss term for constraining the orthogonality of the orthogonal linear layer;

[0091]

[0092] wherein, φkrepresents a feature of the kth element, φkrepresents a predicted distribution of the kth element, θ k ={φ k ,ψ k} represents a set of learnable parameters, φkrepresents a probability distribution of side information;

[0093]

[0094] wherein, I represents a matrix with diagonal values of 1, W enc,j represents a coefficient of an orthogonal linear layer of an orthogonal linear wavelet coding unit of the bypass encoder, W enc,j represents a coefficient of an orthogonal linear layer of an orthogonal linear wavelet decoding unit of the bypass decoder.

[0095] The image compression system provided by the present disclosure can be applied to any existing end-to-end image compression coding network and entropy model network to accelerate the training of the image compression system and improve the training efficiency of the image compression system.

[0096] Figure 2 is a flowchart of an image coding method according to an exemplary embodiment.

[0097] As Figure 2 shown, the present disclosure provides an image coding method, comprising S11 to S16.

[0098] S11, inputting a to-be-compressed image into an encoding network to determine a first feature map of the to-be-compressed image.

[0099] wherein, the encoding network can adopt a spatial frequency modulation encoding network, and the spatial frequency modulation encoding network is an encoding network composed of a main encoder and a bypass encoder.

[0100] The first feature map represents a three-dimensional feature map, and the dimensions of the first feature map include a channel dimension, a height dimension, and a width dimension.

[0101] S12, quantizing each feature value in the first feature map into an integer to determine a second feature map.

[0102] S13, inputting the second feature map into an entropy model, and modeling the distribution of the second feature map as a Gaussian distribution to determine a Gaussian distribution of the second feature map.

[0103] S14, determining a mean and a variance of a predicted Gaussian distribution according to the Gaussian distribution of the second feature map.

[0104] S15, determining a probability estimate of the second feature map according to the mean and the variance of the predicted Gaussian distribution.

[0105] S16, arithmetic coding is performed on the probability estimation of the second feature map to determine the compressed image binary code stream.

[0106] In a possible embodiment, S11 can include S101-S103.

[0107] S101, inputting the image to be compressed into a main encoder to determine a main feature map.

[0108] Wherein the image to be compressed is represented as x, and the main feature map is represented as y main .

[0109] S102, inputting the image to be compressed into a bypass encoder to determine a bypass feature map.

[0110] Wherein the bypass feature map is represented as y aux .

[0111] S103, combining the main feature map and the bypass feature map to determine the first feature map of the image to be compressed.

[0112] Wherein the first feature map is represented as y.

[0113] Wherein specifically, S102, inputting the image to be compressed into a bypass encoder to determine a bypass feature map, can include:

[0114] Inputting the image to be compressed into a first orthogonal linear wavelet coding unit, and in a sequence, the output of each previous orthogonal linear wavelet coding unit is taken as the input of the next orthogonal linear wavelet coding unit, and the last orthogonal linear wavelet coding unit outputs the first bypass feature;

[0115] Wherein the input of each orthogonal linear wavelet coding unit is taken as the input of the discrete wavelet transform of each orthogonal linear wavelet coding unit, the discrete wavelet transform outputs the wavelet transform coding feature map, the output of the discrete wavelet transform is taken as the input of the orthogonal linear layer of each orthogonal linear wavelet coding unit, the orthogonal linear layer outputs the first transform feature, and the output of the orthogonal linear layer is taken as the output of each orthogonal linear wavelet coding unit.

[0116] As an example, the input of the first orthogonal linear wavelet coding unit is the image to be compressed x, the output of the first orthogonal linear wavelet coding unit is x1, and the output of the first orthogonal linear wavelet coding unit x1 is taken as the input of the second orthogonal linear wavelet coding unit, and so on, the input of the jth orthogonal linear wavelet coding unit is the output of the j-1th orthogonal linear wavelet coding unit x j-1 , and the output of the last orthogonal linear wavelet coding unit is the bypass feature map y aux .

[0117] As another example, the following operations are performed within the orthogonal linear wavelet coding unit:

[0118] For the jth orthogonal linear wavelet coding unit, input quantity x j-1 Input the discrete wavelet forward transform, and determine the wavelet transform coding feature map DWT(x j-1 ) through the wavelet forward transform; input the wavelet transform coding feature map DWT(x j-1 ) into the orthogonal linear layer with the coefficient W enc,j , determine the first conversion feature, and take it as the output quantity of the jth orthogonal linear wavelet coding unit. The specific formula is:

[0119] x j = DWT(x j-1 )W enc,j

[0120] By using the bypass encoder and the plurality of orthogonal linear wavelet coding units of the bypass encoder to code the image to be compressed, effective decorrelation is performed, the redundant information in coding is reduced, the code rate overhead in the image compression process is reduced, the image reconstruction quality is improved, and the training process can be accelerated.

[0121] In a possible embodiment, S12 can include S201 to S202.

[0122] S201, input the first feature map into the preset hyper-prior encoder, perform hyper-prior feature extraction, and determine the hyper-prior feature of the first feature map.

[0123] S202, quantize each feature value of the hyper-prior feature of the first feature map to an integer by rounding, and determine the second feature map.

[0124] In a possible embodiment, the entropy model can use a hyper-prior entropy model, and a hyper-prior network is introduced, which includes a hyper-prior encoder and a hyper-prior decoder.

[0125] In a possible embodiment, the image coding method can further include S17 to S20.

[0126] S17, according to the hyper-prior feature in the second feature map, establish a distribution estimation independent of each other.

[0127] S18, use the distribution estimation independent of each other to perform arithmetic coding on the hyper-prior feature in the second feature map, and determine the hyper-prior feature binary code stream.

[0128] Among them, the arithmetic coding can be performed by the hyper-prior encoder of the hyper-prior network in the entropy model.

[0129] S19, arithmetically decoding the super-prior feature binary code stream to determine the recovered super-prior feature.

[0130] The arithmetical decoding can be performed by a super-prior decoder in the super-prior network in the entropy model.

[0131] S20, inputting the recovered super-prior feature into a preset super-decoding neural network to determine the super-prior information.

[0132] Steps S17 to S20 can be performed in the entropy model, and S17 to S20 are performed before step S13.

[0133] In a possible embodiment, S14 can include:

[0134] According to the Gaussian distribution of the second feature map, the mean and variance of the Gaussian distribution of the second feature map are predicted using the super-prior information to determine the predicted mean and variance of the Gaussian distribution.

[0135] In a possible embodiment, S15 can include:

[0136] According to the predicted mean and variance of the Gaussian distribution, a probability estimation of each three-dimensional feature of the second feature map is performed to determine the probability estimation of the second feature map.

[0137] In a possible embodiment, S16 can include S401 to S402.

[0138] S401, arithmetically encoding the probability estimation of the second feature map to obtain a feature binary code stream.

[0139] The feature binary code stream is a feature binary code stream of the three-dimensional feature.

[0140] S402, merging the feature binary code stream and the super-prior feature binary code stream to determine a compressed image binary code stream.

[0141] The super-prior feature binary code stream is a super-prior feature binary code stream of the super-prior feature obtained in step S18.

[0142] In a possible embodiment, an entropy model adopted in an image encoding method can adopt any one of a super-prior entropy model, a spatial auto-regressive entropy model, and a channel auto-regressive entropy model.

[0143] As an example, the entropy model adopts a super-prior entropy model, specifically as follows:

[0144] The hyper-prior network is introduced, and the hyper-prior network includes a hyper-prior encoder and a hyper-prior decoder. The hyper-prior encoder is used to perform arithmetic encoding on the hyper-prior in the second feature map, and the hyper-prior decoder is used to perform Gaussian modeling on each point in the second feature map with a mean of μ and a variance of σ, to determine the Gaussian distribution of the second feature map, and to perform entropy encoding on the probability estimation of the second feature map, to determine the compressed image binary code stream.

[0145] As another example, the entropy model adopts a spatial autoregressive entropy model, which further introduces a spatial autoregressive model on the basis of the hyper-prior entropy model, models the local context correlation of the second feature map in space, and predicts the probability distribution of the feature value of the second feature map.

[0146] As another example, the entropy model adopts a channel autoregressive entropy model, which further introduces a channel conditional model on the basis of the hyper-prior entropy model, models the features of the channels according to the correlation between the channels of the second feature map, and determines the probability distribution of the current channel by predicting the features of the previous channels.

[0147] In a possible embodiment, in an image encoding method, a hyper-prior entropy model and a basic encoding unit based on convolution and a nonlinear unit can be used, and a basic encoding unit of a Transformer and an autoregressive-based entropy model can also be used, all of which can achieve the purpose of image encoding of the present disclosure.

[0148] Figure 3 is a flowchart of an image decoding method according to an example embodiment.

[0149] As Figure 3 shown, the present disclosure also provides an image decoding method, which can decode the compressed image binary code stream obtained by the above-mentioned encoding method, and includes S21 to S24.

[0150] S21, obtaining the compressed image binary code stream.

[0151] S22, inputting the compressed image binary code stream into the entropy model to determine the mean and variance of the predicted Gaussian distribution of the second feature map.

[0152] S23, performing arithmetic decoding according to the mean and variance of the predicted Gaussian distribution of the second feature map to determine the second feature map.

[0153] S24, inputting the second feature map into the decoding network to determine the reconstructed image.

[0154] In a possible embodiment, S22 can include S221.

[0155] S221, input the compressed image binary code stream into the entropy model, code stream decomposition is carried out on the compressed image binary code stream, and feature binary code stream and hyper-prior feature binary code stream are obtained.

[0156] In a possible embodiment, S22 can further include S222-S223.

[0157] S222, arithmetic decoding is performed on the hyper-prior feature binary code stream, and recovered hyper-prior features are obtained.

[0158] S223, the recovered hyper-prior features are input into a hyper-decoding neural network, and hyper-prior information is obtained.

[0159] Steps S221-S223 can be performed in the entropy model.

[0160] In a possible embodiment, S22 can further include S224.

[0161] S224, the hyper-prior information is used to predict the mean and variance of the Gaussian distribution of the predicted second feature map.

[0162] The Gaussian distribution of the predicted second feature map is a Gaussian model of each feature in the predicted second feature map.

[0163] The hyper-prior information used in step S224 is the hyper-prior information obtained in step S223.

[0164] In a possible embodiment, S23 can include:

[0165] According to the mean and variance of the Gaussian distribution of the predicted second feature map, the feature binary code stream is arithmetically decoded, and a second feature map is determined, the second feature map containing quantized three-dimensional features.

[0166] The decoding network used in the present disclosure is a null frequency modulation decoding network, which includes a main decoder and a bypass decoder.

[0167] In a possible embodiment, S24 can include S241-S243.

[0168] S241, the second feature map is input into the main decoder, and a main decoded image is determined.

[0169] The main decoded image is represented as

[0170] S242, the second feature map is input into the bypass decoder, and a bypass decoded image is determined.

[0171] The bypass decoded image is represented as

[0172] S243, combining the main decoded image and the bypass decoded image to determine the reconstructed image.

[0173] wherein the reconstructed image is represented as

[0174] Specifically, S242, inputting the second feature map into the bypass decoder to determine the bypass decoded image, can include:

[0175] inputting the second feature map into the first orthogonal linear wavelet decoding unit, and according to the order, taking the output of each previous orthogonal linear wavelet decoding unit as the input of the next orthogonal linear wavelet decoding unit, and taking the output of the last orthogonal linear wavelet decoding unit as the bypass decoded image;

[0176] wherein the input of each orthogonal linear wavelet decoding unit is taken as the input of the inverse discrete wavelet transform of each orthogonal linear wavelet decoding unit, the inverse discrete wavelet transform outputs the discrete wavelet feature, the output of the inverse discrete wavelet transform is taken as the input of the orthogonal linear layer of each orthogonal linear wavelet decoding unit, the orthogonal linear layer outputs the second transformed feature, and the output of the orthogonal linear layer is taken as the output of each orthogonal linear wavelet decoding unit.

[0177] As an example, the input of the first orthogonal linear wavelet decoding unit is the second feature map, the input of the first orthogonal linear wavelet decoding unit is y1, and the output of the first orthogonal linear wavelet decoding unit y1 is taken as the input of the second orthogonal linear wavelet decoding unit, and so on, the input of the jth orthogonal linear wavelet decoding unit is the output of the (j-1)th orthogonal linear wavelet decoding unit y j-1 , and the output of the last orthogonal linear wavelet decoding unit is the bypass decoded image

[0178] As another example, the following operations are performed in the orthogonal linear wavelet decoding unit:

[0179] For the jth orthogonal linear wavelet decoding unit, the input y j-1 is input into the inverse discrete wavelet transform to determine the discrete wavelet feature through the inverse wavelet transform; the discrete wavelet feature is input into the orthogonal linear layer with the coefficient W dec,j to determine the second transformed feature, which is taken as the output of the jth orthogonal linear wavelet decoding unit.

[0180] By using the main decoder and the bypass decoder to obtain the reconstructed image, the reconstruction quality of the reconstructed image is effectively improved.

[0181] In a possible embodiment, the present disclosure further provides an image compression method, comprising:

[0182] obtaining a compressed image;

[0183] The image encoding method provided by the present disclosure is used to encode the to-be-compressed image, and a compressed image binary code stream is obtained.

[0184] The image decoding method provided by the present disclosure is used to decode the compressed image binary code stream, and a reconstructed image is obtained.

[0185] Figure 4 A flowchart of an image compression method according to an exemplary embodiment is shown.

[0186] As shown in Figure 4 The present disclosure also provides an image compression method, which includes S31 to S45.

[0187] S31, a to-be-compressed image is obtained.

[0188] S32, the to-be-compressed image is input into an encoding network, and a first feature map of the to-be-compressed image is determined.

[0189] S33, each feature value in the first feature map is quantized into an integer, and a second feature map is determined.

[0190] S34, a distribution estimation independent of each other between features is established according to a hyper-prior feature in the second feature map.

[0191] S35, the hyper-prior feature in the second feature map is arithmetically encoded using the distribution estimation independent of each other between features, and a hyper-prior feature binary code stream is determined.

[0192] S36, the hyper-prior feature binary code stream is arithmetically decoded, and a recovered hyper-prior feature is determined.

[0193] S37, the recovered hyper-prior feature is input into a preset hyper-decoding neural network, and hyper-prior information is determined.

[0194] S38, the second feature map is input into an entropy model, and a distribution of the second feature map is modeled as a Gaussian distribution, and a Gaussian distribution of the second feature map is determined.

[0195] S39, a mean value and a variance of a predicted Gaussian distribution are determined according to the Gaussian distribution of the second feature map.

[0196] The mean value and the variance of the predicted Gaussian distribution are determined according to the Gaussian distribution of the second feature map using the hyper-prior information to predict the mean value and the variance of the Gaussian distribution of the second feature map.

[0197] S40, a probability estimation of the second feature map is determined according to the mean value and the variance of the predicted Gaussian distribution.

[0198] The probability estimation of the second feature map is determined according to the mean and variance of the predicted Gaussian distribution of the second feature map.

[0199] S41, the probability estimation of the second feature map is arithmetically encoded to determine the compressed image binary code stream.

[0200] The probability estimation of the second feature map is arithmetically encoded to obtain a feature binary code stream, and the feature binary code stream and the hyper-prior feature binary code stream are merged to determine the compressed image binary code stream.

[0201] S42, the compressed image binary code stream is obtained.

[0202] S43, the compressed image binary code stream is input into an entropy model to determine the mean and variance of the predicted Gaussian distribution of the second feature map.

[0203] The compressed image binary code stream is input into the entropy model, the compressed image binary code stream is code stream decomposed to obtain the feature binary code stream and the hyper-prior feature binary code stream, the hyper-prior feature binary code stream is arithmetically decoded to obtain the recovered hyper-prior feature, the recovered hyper-prior feature is input into a hyper-decoding neural network to obtain the hyper-prior information, and the hyper-prior information is used to predict the mean and variance of the predicted Gaussian distribution of the second feature map.

[0204] S44, the second feature map is determined according to the arithmetical decoding of the mean and variance of the predicted Gaussian distribution of the second feature map.

[0205] The second feature map is determined according to the arithmetical decoding of the mean and variance of the predicted Gaussian distribution of the second feature map, and the second feature map contains quantized three-dimensional features.

[0206] S45, the second feature map is input into a decoding network to determine the reconstructed image.

[0207] The second feature map is input into a main decoder to determine a main decoded image, the second feature map is input into a bypass decoder to determine a bypass decoded image, and the main decoded image and the bypass decoded image are combined to determine the reconstructed image.

[0208] The image compression method provided by the disclosure adopts a trained image compression system, can efficiently perform decorrelation decomposition on image information, reduce redundant information, reduce image compression code rate overhead, and improve image reconstruction quality.

[0209] Figure 5 is a block diagram of an image compression system according to an example embodiment.

[0210] As Figure 5 shown in a possible embodiment, the present disclosure provides an image compression system 1000, comprising:

[0211] An acquisition module 1100 is configured to acquire an image to be compressed.

[0212] An encoder 1200 is configured to encode the image to be compressed by using the image encoding method provided by the present disclosure, and obtain a compressed image binary code stream.

[0213] A decoder 1300 is configured to decode the compressed image binary code stream by using the image decoding method provided by the present disclosure, and determine a reconstructed image.

[0214] Regarding the embodiments of the above system, the specific manner in which each device performs the operation has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0215] In a possible embodiment, 24 images with a resolution of 512x768 in the Kodak test set are compressed by using the image compression system 1000 provided by the present disclosure, comprising:

[0216] An acquisition module is configured to acquire the images to be compressed in the Kodak test set in sequence.

[0217] An encoder is configured to encode the image to be compressed according to the above steps S31 to S35, and determine a compressed image binary code stream.

[0218] Among them, the channel number of the three-dimensional feature of the first feature map of the image to be compressed is 230, the height is 32, and the width is 48.

[0219] A decoder is configured to decode the compressed image binary code stream according to the above steps S36 to S45, and determine a reconstructed image.

[0220] Among them, the channel number of the three-dimensional feature of the decoded reconstructed image is 230, the height is 32, and the width is 48.

[0221] Figure 6 is a flow chart of an image compression system training method according to an example embodiment.

[0222] As Figure 6 shown in a possible embodiment, one million natural images in the ImageNet-train dataset are used to train the image compression system provided by the present disclosure, and the trained image compression system can be used for encoding and decoding images, comprising S51 to S59.

[0223] S51, three-dimensional feature extraction: the images in the training set are randomly cropped into 256x256 resolution size images, input into the encoding network composed of the main encoder and the bypass encoder to obtain three-dimensional features,

[0224] wherein the number of three-dimensional feature channels is 192, and the height and width are both 16.

[0225] Of course, the image resolution size and the three-dimensional feature parameters described above in the embodiment can be adjusted according to actual conditions, and are not limited to the convolution module cascade in the step. The parameters in the following steps are also the same.

[0226] Correspondingly, the main encoder can also use other network structures, such as a cascade of Transformer blocks and residual blocks, and is not limited to the convolution module cascade in the step.

[0227] S52, super-prior feature extraction and noise addition: inputting the three-dimensional features into the super-prior encoding neural network to obtain super-prior features, and adding -0.5 to 0.5 uniformly distributed noise to the super-prior features to simulate actual quantization in order to avoid the gradient being zero everywhere caused by quantization.

[0228] wherein: the super-prior encoding neural network is composed of three convolution layers in cascade; of course, the super-prior encoding neural network can also be other network structure forms, such as a cascade of three residual modules, and is not limited to the super-prior encoding neural network in the step.

[0229] S53, super-prior feature code rate estimation: a distribution estimation independent of each other between features is established for the super-prior features, and a super-prior feature code rate estimation is obtained using the distribution estimation.

[0230] S54, super-prior information acquisition: inputting the super-prior features with added noise into the super-prior decoding neural network to obtain super-prior information, wherein: the super-prior decoding neural network is composed of three convolution layers in cascade.

[0231] S55, feature noise addition: -0.5 to 0.5 uniformly distributed noise is added to the three-dimensional features to simulate actual quantization in order to avoid the gradient being zero everywhere caused by quantization.

[0232] S56, Gaussian distribution probability modeling: the structure of this module is exactly the same as that of the probability entropy modeling in the embodiment shown in Figure 2 ; the mean and variance of the Gaussian distribution are predicted using the super-prior information; of course, the entropy model network can also be other forms, such as a spatial context autoregressive model and a reconstruction context autoregressive model.

[0233] S57, feature code rate estimation: the feature code rate estimation is obtained according to the predicted mean and variance of the Gaussian distribution.

[0234] S58, total code rate estimation: add the super-prior feature code rate estimation and the feature code rate estimation to obtain a total code rate estimation.

[0235] S59, image reconstruction: input the three-dimensional features after adding noise into an encoding network composed of a main encoder and a bypass encoder to obtain a reconstructed image.

[0236] In a possible embodiment, the four image compression systems provided by the present disclosure are trained using the ImageNet dataset, and λ is set to 0.0018, 0.0035, 0.0067 and 0.0130 respectively, that is, corresponding to four image compression systems with different code rates. The Adam optimizer is used for 1,000,000 steps of optimization, and the learning rate is set to 10 -4 in the first 900,000 steps and is reduced to 10 -5 in the last 100,000 steps.

[0237] The four image compression systems obtained by the above training are used to compress 24 images with a resolution of 512*768 in the Kodak test set. The average code rate of the compressed images is 0.1284 bpp, 0.1983 bpp, 0.2963 bpp and 0.4332 bpp respectively, and the average distortion PSNR of the reconstructed images and the original images is 29.650, 31.142, 32.712 and 34.430 respectively.

[0238] Compared with the method using the benchmark neural network, the image compression system provided by the present disclosure has better rate-distortion performance and faster training convergence speed. Specifically, under the premise of the same reconstruction quality, the coding code rate overhead can be reduced by 1-2%, and the training time can be saved by 50%.

[0239] Based on the same idea, in another embodiment of the present disclosure, a terminal is also provided, which includes a memory, a processor and a computer program stored in the memory and capable of running on the processor. The processor is configured to execute the image encoding method, or the image decoding method, or the image compression method when executing the program.

[0240] Optionally, the memory is configured to store a program; the memory can include volatile memory (e.g., random access memory (RAM), such as static random access memory (SRAM), Double Data Rate SDRAM (DDR SDRAM), etc.), and / or non-volatile memory (e.g., flash memory). The memory is configured to store computer programs (e.g., application programs, functional modules, etc. for implementing the above-described methods), computer instructions, etc. The computer programs, computer instructions, etc. described above can be stored in one or more memories in a distributed manner. Moreover, the computer programs, computer instructions, data, etc. described above can be invoked by the processor.

[0241] The computer programs, computer instructions, etc. described above can be stored in one or more memories in a distributed manner. Moreover, the computer programs, computer instructions, data, etc. described above can be invoked by the processor.

[0242] The processor is configured to execute the computer programs stored in the memory to implement each step in the methods described above in the embodiments. Details can be referred to the related descriptions in the method embodiments above.

[0243] The processor and the memory can be independent structures, or can be integrated into an integrated structure. When the processor and the memory are independent structures, the memory and the processor can be coupled and connected through a bus.

[0244] In the embodiments of the present disclosure, a non-transitory computer-readable storage medium is also provided, which stores a computer program. The computer program is executed by a processor to implement the steps of the image encoding method, or the image decoding method, or the image compression method in any of the above-described embodiments.

[0245] Those skilled in the art should understand that the embodiments of the present disclosure can be provided as a method, a system, or a computer program product. Therefore, the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer-usable program code.

[0246] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the one or more flowcharts and / or blocks

[0247] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the one or more flowcharts and / or blocks

[0248] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the one or more flowcharts and / or blocks

[0249] While the preferred embodiments of the disclosure have been described, additional variations and modifications can be employed by those skilled in the art once armed with the concepts reflected in the disclosure without departing from the scope of the disclosure. Accordingly, the appended claims are intended to embrace all such additional variations and modifications as fall within the scope of the disclosure.

[0250] It will be apparent to those skilled in the art that various modifications and variations can be made to the present disclosure without departing from the spirit or scope of the disclosure. Thus, it is intended that the present disclosure cover the modifications and variations of this disclosure provided such modifications and variations come within the scope of the appended claims and their equivalents.

Claims

1. An image compression system characterized by, The application relates to an image compression method and device. The encoding network comprises a main encoder and a bypass encoder which are connected in parallel, and the encoding network is used for encoding an image to be compressed, and the bypass encoder adopts orthogonal linear constraint, wherein the main encoder adopts an encoding network of an end-to-end image compression system, and the main encoder comprises a plurality of convolution blocks or a plurality of Transformer blocks, and the bypass encoder comprises a plurality of orthogonal linear wavelet coding units, and each orthogonal linear wavelet coding unit comprises a cascaded discrete wavelet forward transform and an orthogonal linear layer. The decoding network comprises a main decoder and a bypass decoder which are connected in parallel, and the decoding network is used for decoding a compressed image binary code stream, and the bypass decoder adopts orthogonal linear constraint, wherein the main decoder adopts a decoding network of an end-to-end image compression system, and the main decoder comprises a plurality of convolution blocks or a plurality of Transformer blocks, and the bypass decoder comprises a plurality of orthogonal linear wavelet decoding units, and each orthogonal linear wavelet decoding unit comprises a cascaded discrete wavelet inverse transform and an orthogonal linear layer. The entropy model is used for predicting the mean and variance of a Gaussian distribution. The training loss function of the encoding network, the decoding network and the entropy model is as follows:

2. The image compression system of claim 1, wherein, The application relates to an image compression method and device. wherein, denotes a distortion, x denotes the image to be compressed, denotes a reconstructed image, λ1 denotes a first balancing parameter, λ2 denotes a second balancing parameter, R1 denotes a code rate, denotes a loss term for constraining the orthogonality of the orthogonal linear layers; where, φkdenotes the feature of the kth element, φkdenotes the predicted distribution of the kth element, k = {φ k , ψ k} denotes a set of learnable parameters, φkdenotes the predicted distribution of the kth element; where I denotes a matrix with diagonal values of 1, W enc,j denotes coefficients of the orthogonal linear layer of the orthogonal linear wavelet encoding unit of the bypass encoder, W enc,j denotes coefficients of the orthogonal linear layer of the orthogonal linear wavelet decoding unit of the bypass decoder.

3. An image coding method characterized by, The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device.

4. The method of claim 3, wherein, The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device.

5. The method of claim 4, wherein, The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an image compression method and device. The application relates to an Wherein, the input of each of the orthogonal linear wavelet coding units is the input of a discrete wavelet forward transform of each of the orthogonal linear wavelet coding units, the discrete wavelet forward transform outputs a wavelet transform coding feature map, the output of the discrete wavelet forward transform is the input of an orthogonal linear layer of each of the orthogonal linear wavelet coding units, the orthogonal linear layer outputs a first transform feature, the output of the orthogonal linear layer is the output of each of the orthogonal linear wavelet coding units.

6. An image decoding method characterized by, The compressed image binary code stream obtained in claim 5 is decoded, comprising: Obtaining a compressed image binary code stream; Inputting the compressed image binary code stream into the entropy model in claim 1 to determine the mean and variance of the Gaussian distribution of the predicted second feature map; According to the mean and variance of the Gaussian distribution of the predicted second feature map, the second feature map is determined by arithmetic decoding; Inputting the second feature map into the decoding network in claim 1 to determine a reconstructed image.

7. The method of claim 6, wherein, The second feature map is inputted into the decoding network to determine a reconstructed image, comprising: Inputting the second feature map into a main decoder to determine a main decoded image; Inputting the second feature map into a bypass decoder to determine a bypass decoded image; Combining the main decoded image and the bypass decoded image to determine the reconstructed image.

8. The method of claim 6, wherein, The second feature map is inputted into the bypass decoder to determine a bypass decoded image, comprising: Inputting the second feature map into a first orthogonal linear wavelet decoding unit, in a sequence, the output of each previous orthogonal linear wavelet decoding unit is the input of a next orthogonal linear wavelet decoding unit, and the last orthogonal linear wavelet decoding unit outputs the bypass decoded image; Wherein, the input of each of the orthogonal linear wavelet decoding units is the input of a discrete wavelet inverse transform of each of the orthogonal linear wavelet decoding units, the discrete wavelet inverse transform outputs a discrete wavelet feature, the output of the discrete wavelet inverse transform is the input of an orthogonal linear layer of each of the orthogonal linear wavelet decoding units, the orthogonal linear layer outputs a second transform feature, and the output of the orthogonal linear layer is the output of each of the orthogonal linear wavelet decoding units.

9. An image compression method characterized by, Comprising: Obtaining an image to be compressed; Encoding the image to be compressed based on the image encoding method in any one of claims 3-5 to obtain a compressed image binary code stream; Decoding the compressed image binary code stream based on the image decoding method in any one of claims 6-8 to obtain a reconstructed image.

10. A terminal, characterized by comprising: A device comprising a memory and a processor, the memory storing a program executed by the processor, and the processor can execute any one of the following methods when executing the program: The image encoding method in any one of claims 3-5, or The image decoding method in any one of claims 6-8, or The image compression method in claim 9.

11. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, the image encoding method in any one of claims 3-5, or the image decoding method in any one of claims 6-8, or the image compression method in claim 9 can be executed.

Images