Encoding and decoding method and system with controllable entropy decoding complexity

Abstract

The present invention discloses a coding and decoding method and system with controllable entropy decoding complexity. The method uses a highly scalable and complexity-controllable entropy decoding scheme, so that the decoding of semantically structured code streams can support the complexity requirements of any external setting, thereby adapting to intelligent analysis tasks in different application scenarios, and improving the versatility and flexibility of the semantically structured image coding and decoding scheme. At the same time, each target in the encoding process can adopt entropy coding reference dependency relationships of different complexities for probabilistic modeling, further enhancing the flexibility and scalability of the coding and decoding scheme, making it more suitable for real application scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of image compression coding, and in particular to a coding and decoding method and system with controllable entropy decoding complexity. Background Art

[0002] Existing image compression technologies primarily target human vision. However, with the rapid development of deep learning, machine intelligence analysis is increasingly playing a vital role in various areas of human production and life, such as smart industry and smart cities. Traditional compression methods targeting the human eye require all image information to be compressed and transmitted. At the intelligent task analysis end, the entire image must be decoded to obtain a complete image before being fed into the subsequent intelligent task analysis model.

[0003] To more efficiently support human-machine hybrid intelligent applications, existing methods have proposed the concept of semantically structured image code streams. For example, Solution 1, a Chinese invention patent with authorization publication number CN110225341 B, entitled "A Task-Driven Code Stream Structured Image Coding Method," incorporates a region decision network and alignment module for object detection. Based on compressed features, bounding boxes are extracted for the object's location, and the features are spatially segmented based on these bounding boxes. The segmented features are then sequentially entropy-encoded to form a structured code stream.

[0004] However, in Solution 1, the entropy decoding part uses a hyper-prior-based approach to model probability estimation. Although this entropy decoding method is fast, it has low coding efficiency and cannot more efficiently support downstream intelligent analysis tasks. Although it is also possible to use a spatial autoregressive approach (Minnen D, Ballé J, Toderici G D. Joint autoregressive and hierarchical priors for learned image compression [J]. Advances in neural information processing systems, 2018, 31.) for probability estimation to achieve higher compression performance, this method's entropy decoding speed is very slow and cannot meet the needs of practical applications. Summary of the Invention

[0005] The purpose of the present invention is to provide a coding and decoding method and system with controllable entropy decoding complexity, which can support any predefined complexity requirements, adapt to intelligent analysis tasks in different application scenarios while introducing extremely low bit rates, and improve the versatility and flexibility of the semantically structured image coding method.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A coding and decoding method with controllable entropy decoding complexity, comprising:

[0008] The encoding part: performs target detection on the input image to obtain target detection results; uses the transformation operation of the deep image encoder to obtain image features of the input image, performs super-prior transformation and quantization on the image features to obtain discrete super-prior features, performs super-prior inverse transformation on the discrete super-prior features to obtain an overall probability distribution, and separately encodes the size information of the input image, the discrete super-prior features, and the target detection results, and then splices them to obtain code stream header information; quantizes the image features to obtain overall discrete image features, groups the overall discrete image features based on the target detection results to obtain discrete image features of each target, determines the autoregressive reference relationship corresponding to each target based on a predefined autoregressive reference relationship set and a selected decoding complexity, determines the probability distribution of each target based on the autoregressive reference relationship of each target and the overall probability distribution, and performs entropy coding on the discrete image features of the corresponding target to obtain a texture part code stream corresponding to each target, and encodes the index of the autoregressive reference relationship of each target and combines it with the texture part code stream of the corresponding target to form a target corresponding code stream; all target corresponding code streams and the code stream header information constitute a semantically structured code stream;

[0009] Decoding part: Entropy decoding is performed on the bitstream header information in the semantically structured bitstream to obtain the size information of the input image, the overall discrete super-prior features and the target detection results; the overall discrete super-prior features are inversely transformed to obtain the overall probability distribution; for each target, the corresponding discrete image features are entropy decoded from the target corresponding bitstream based on the overall probability distribution, and the discrete image features corresponding to all targets are spliced ​​according to the target detection results to obtain the overall discrete image features; then, the overall discrete image features are combined with the size information of the input image, and the reconstructed image is obtained through the inverse transformation operation of the deep image decoder.

[0010] A coding and decoding system with controllable entropy decoding complexity, comprising:

[0011] The encoding unit is used to execute the encoding part, which includes: performing target detection on the input image to obtain the target detection result; using the transformation operation of the deep image encoder to obtain the image features of the input image, performing super-prior transformation and quantization on the image features to obtain discrete super-prior features, performing super-prior inverse transformation on the discrete super-prior features to obtain the overall probability distribution, and encoding the size information of the input image, the discrete super-prior features and the target detection results respectively, and then splicing them to obtain the code stream header information; quantizing the image features to obtain the overall discrete image features, combining the target detection results to obtain the overall discrete The image features are grouped to obtain discrete image features of each target. The autoregressive reference relationship corresponding to each target is determined based on a predefined set of autoregressive reference relationships and a selected decoding complexity. The probability distribution of each target is determined based on the autoregressive reference relationship of each target and the overall probability distribution. The discrete image features of the corresponding targets are entropy coded to obtain a texture portion bitstream corresponding to each target. The index of the autoregressive reference relationship of each target is encoded and then combined with the texture portion bitstream of the corresponding target to form a target corresponding bitstream. All target corresponding bitstreams and the bitstream header information constitute a semantically structured bitstream.

[0012] A decoding unit is used to execute the decoding part, which includes: entropy decoding the code stream header information in the semantically structured code stream to obtain the size information of the input image, the overall discrete super prior features and the target detection results; performing a super prior inverse transform on the overall discrete super prior features to obtain the overall probability distribution; for each target, entropy decoding the corresponding discrete image features from the target corresponding code stream in combination with the overall probability distribution, splicing the discrete image features corresponding to all targets according to the target detection results to obtain the overall discrete image features; then combining the overall discrete image features with the size information of the input image, and obtaining the reconstructed image through the inverse transform operation of the deep image decoder.

[0013] A processing device comprising: one or more processors; a memory for storing one or more programs;

[0014] When the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method.

[0015] A readable storage medium stores a computer program, which implements the aforementioned method when the computer program is executed by a processor.

[0016] It can be seen from the technical solution provided by the present invention that a highly scalable and complexity-controllable entropy decoding scheme is used, so that the decoding of the semantically structured code stream can support the complexity requirements of any external setting, thereby adapting to intelligent analysis tasks in different application scenarios, and improving the versatility and flexibility of the semantically structured image encoding and decoding scheme; at the same time, each target in the encoding process can adopt entropy coding reference dependencies of different complexities for probabilistic modeling, further enhancing the flexibility and scalability of the encoding and decoding scheme, making it more suitable for real application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flowchart of a coding and decoding method with controllable entropy decoding complexity provided by an embodiment of the present invention;

[0019] Figure 2 A schematic diagram of a coding and decoding system with controllable entropy decoding complexity provided by an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of a processing device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0022] First, the following terms may be used in this article:

[0023] The term “and / or” means that either or both of them can be realized at the same time. For example, X and / or Y includes both “X” or “Y” and “X and Y”.

[0024] The terms "include," "comprises," "contains," "has," or other similar expressions should be interpreted as non-exclusive. For example, "including certain technical features (such as raw materials, components, ingredients, carriers, dosage forms, materials, dimensions, parts, components, mechanisms, devices, steps, procedures, methods, reaction conditions, processing conditions, parameters, algorithms, signals, data, products, or manufactured articles, etc.) should be interpreted as including not only the technical features explicitly listed, but also other technical features known in the art that are not explicitly listed.

[0025] The term "consisting of" excludes any technical features not explicitly listed. If used in a claim, this term renders the claim closed, excluding any technical features other than those explicitly listed, except for conventional impurities associated with them. If this term appears only in a clause of a claim, it limits only the elements explicitly listed in that clause; elements listed in other clauses are not excluded from the claim as a whole.

[0026] The following describes in detail a coding and decoding scheme with controllable entropy decoding complexity provided by the present invention. Any information not described in detail in the embodiments of the present invention is prior art known to those skilled in the art. For any unspecified conditions in the embodiments of the present invention, the conditions conventional in the art or recommended by the manufacturer shall prevail.

[0027] Example 1

[0028] The present invention provides a coding and decoding method with controllable entropy decoding complexity. This method improves semantically structured image coding methods to achieve controllable coding and decoding complexity. Specifically, it introduces a function that controls entropy decoding complexity based on semantically structured coding methods. The core technology is that different reference dependencies can be selected for each target throughout the coding and decoding process to achieve different complexities, thereby achieving a balance between decoding complexity and coding efficiency. Figure 1 The main process of the encoding and decoding method is demonstrated, which mainly includes two parts: encoding and decoding.

[0029] 1. Coding part.

[0030] 1. Perform target detection on the input image to obtain the target detection result.

[0031] In the embodiment of the present invention, target detection is performed on the input image by using a target detection model to obtain the number of targets, the position and category of each target (target detection result); wherein the position and category of the kth target are expressed as: {a k ,b k ,h k ,w k ,clsk}, a k ,b k ,h k ,w k ,cls k They represent the horizontal position of the upper left corner coordinate, the vertical position of the upper left corner coordinate, the height, the width and the category of the k-th target respectively.

[0032] In the embodiment of the present invention, the target detection model can be implemented by an existing network model, so it is not described in detail.

[0033] 2. Use the transformation operation of the deep image encoder to obtain the image features of the input image.

[0034] In the embodiment of the present invention, the size of the input image x is recorded as H×W×C, where H and W represent the height and width of the image respectively, and C is the number of channels (all channels are encoded and decoded at the same time). The image feature y is obtained by the transformation operation, and the size is C y is the number of channels, and B is the downsampling multiple after the transformation is completed.

[0035] In the embodiment of the present invention, the depth image encoder can be implemented by an existing network model, so it is not described in detail. The transformation operations involved can be nonlinear transformation operations.

[0036] 3. Perform a super-prior transformation and quantization on the image features to obtain discrete super-prior features, perform a super-prior inverse transformation on the discrete super-prior features to obtain the overall probability distribution, and encode the size information of the input image, the discrete super-prior features and the target detection results respectively, and then splice them to obtain the code stream header information.

[0037] like Figure 1 As shown in the right part, the image feature y is transformed into a super-prior feature z, and the super-prior feature z is quantized to obtain a discrete super-prior feature After that, the height H and width W of the input image (the size information of the input image), the discrete hyper-prior feature and target detection results are encoded separately, and then all the information obtained by encoding is spliced ​​in sequence to obtain the code stream header information; wherein, the discrete super prior features are Entropy coding is performed, and other information can be coded using conventional coding. When encoding the target detection results separately, the number of targets, the location of each target, and the category are encoded separately. The syntax structure of the codestream header information is defined as shown in Table 1.

[0038] Table 1: Syntax structure of codestream header information

[0039]

[0040] Among them: image_height_minus1 represents the height H of the image; image_width_minus1 represents the width W of the image; side_information_length represents the quantized super prior feature The corresponding bitstream length; bounding_boxes_numbers indicates the number of targets in the image; bounding_box_x, bounding_box_y, bounding_box_h, bounding_box_w, and bounding_box_category respectively represent the horizontal coordinate of the upper left corner, the vertical coordinate of the upper left corner, the height, the width, and the category information of the current target. Only an example of a single target is provided here. The above five information of multiple targets are grouped together and arranged in sequence; u indicates an unsigned data type. Taking u(32) as an example, it indicates that the length of the bitstream segment is 32 bits.

[0041] In the embodiment of the present invention, the code stream length is used for subsequent decoding. The principle is that during actual entropy coding, the length of the encoded data cannot be known, and during decoding, a code stream of the corresponding length must be read before decoding. Therefore, before storing the entropy coded code stream, the corresponding code stream length must be added so that the decoding end can correctly decode the data.

[0042] In addition, the quantized super-prior features need to be inversely transformed to obtain the overall probability distribution And provide it for group feature entropy coding.

[0043] Those skilled in the art will appreciate that there are many ways to implement encoding, and entropy encoding is a specific way of encoding. That is, when described as encoding, it means that entropy encoding can be selected or other encoding methods can be selected.

[0044] 4. Quantize the image features to obtain overall discrete image features, group the overall discrete image features in combination with the target detection results to obtain discrete image features of each target, determine the autoregressive reference relationship corresponding to each target based on a predefined autoregressive reference relationship set and a selected decoding complexity, determine the probability distribution of each target based on the autoregressive reference relationship of each target and the overall probability distribution, and perform entropy coding on the discrete image features of the corresponding target to obtain a texture part code stream corresponding to each target, and encode the index of the autoregressive reference relationship of each target and combine it with the texture part code stream of the corresponding target to form a target corresponding code stream.

[0045] In the embodiment of the present invention, the overall probability distribution refers to the probability distribution of the overall discrete image features, and the probability distribution of all targets can be obtained through the overall probability distribution.

[0046] like Figure 1 As shown in the middle part, the image feature y is quantized to obtain the overall discrete image feature According to the previous target detection results, for each target, assuming that the target index is k, in the discrete image features The corresponding discrete image features are obtained by dividing the spatial dimension The predefined autoregressive reference relation set is denoted as R = {R1, R2, ..., R M}, R i Represents the i-th autoregressive reference relation. Different autoregressive reference relations correspond to different complexities. i is the index of the autoregressive reference relation, i = 1, 2, ..., M, and M is the number of autoregressive reference relations. For the k-th target, the selected decoding complexity is recorded as T k , k=1,2,…,K, K is the number of targets; for the kth target, combined with the predefined autoregressive reference relationship set R and the selected decoding complexity T k , determine the autoregressive reference relationship corresponding to the complexity as R k,i , where R k,i ∈R, k,i is the index of the autoregressive reference relation of the kth target, indicating the correspondence between the target and the autoregressive reference relation. Table 2 shows examples of reference relation indices and corresponding autoregressive reference relations.

[0047] Table 2: Examples of autoregressive reference relationship indices and corresponding autoregressive reference relationships

[0048]

[0049] In the embodiment of the present invention, decoding complexity can be understood as the time required for decoding. Generally speaking, the higher the complexity, the longer the decoding time, and the higher the coding efficiency; while the lower the complexity, the shorter the decoding time, but the lower the coding efficiency. In the embodiment of the present invention, the complexity of the coding can be customized according to the needs. The autoregressive reference relationship affects the length of the decoding time, that is, the size of the complexity. The pre-defined autoregressive reference relationship set is a set of all reference relationships. Therefore, the appropriate reference relationship can be selected from the pre-defined autoregressive reference relationship set according to the required complexity, thereby achieving an ideal complexity-coding efficiency balance.

[0050] In the embodiment of the present invention, the process of entropy coding for each target is the same, and the following is an introduction taking the kth target as an example. The entropy coding needs to use the overall probability distribution obtained above. For the kth target, the autoregressive model is combined with the autoregressive reference relation R of the kth target k,i and the overall probability distribution Discrete image features for the kth target Perform probability modeling to obtain probability distribution Through the second entropy model, combined with probability distribution Discrete image features for the kth target Perform entropy coding to obtain the texture part code stream corresponding to the k-th target; at the same time, the autoregressive reference relationship r of the k-th target is also k,i The index k,i of the target is encoded and combined with the texture part code stream to form the kth target corresponding code stream. The syntax structure definition of each target corresponding code stream is shown in Table 3.

[0051] Table 3: Syntax structure of target corresponding code stream

[0052]

[0053] Wherein, entropy_coding_context_index represents the index of the autoregressive reference relationship, and object_texture_length_minus1 represents the length of the bitstream corresponding to the current target texture portion. It should be noted that the above syntax structure only provides examples related to a single target.

[0054] In the embodiment of the present invention, all target corresponding code streams and the code stream header information constitute a semantically structured code stream.

[0055] 2. Decoding part.

[0056] 1. Perform entropy decoding on the header information of the semantically structured code stream to obtain the size information (height H and width W) of the input image and the overall discrete hyper-prior features. and target detection results.

[0057] 2. For the overall discrete hyper-prior feature Perform super prior inverse transformation to obtain the overall probability distribution

[0058] In the embodiment of the present invention, the aforementioned super-a priori transformation and the super-a priori inverse transformation here can both be implemented through the existing super-a priori model, and thus will not be described in detail.

[0059] 3. For each target, combine the overall probability distribution The corresponding discrete image features are entropy decoded from the target corresponding code stream, and the overall discrete image features are obtained by splicing the discrete image features corresponding to all targets according to the target detection results.

[0060] Taking the kth target as an example, we first decode the autoregressive reference relationship index k,i from the target corresponding code stream, and use it to determine the autoregressive reference relationship R k,i ; Combined with the autoregressive reference relation R k,i and the overall probability distribution Perform autoregressive entropy decoding to obtain the discrete image features corresponding to the kth target Specifically, we need to combine R with the autoregressive model k,i and Perform probability modeling to obtain the k-th target corresponding to Then by utilizing Perform entropy decoding on the texture part of the code stream corresponding to the kth target to obtain the corresponding discrete image features All targets refer to the above method to obtain the discrete image features corresponding to all targets.

[0061] As mentioned before, the target detection result contains the location of each target. Therefore, the discrete image features corresponding to all targets are spliced ​​together according to the location to obtain the overall discrete image features.

[0062] In the embodiment of the present invention, lossless entropy coding and entropy decoding are used. The features obtained by entropy coding and entropy decoding are exactly the same, so the same symbols are used for representation.

[0063] Considering the overall probability distribution obtained by the decoding part Same as the encoding part above, and in the encoding and decoding part, the overall probability distribution is used Get the probability distribution corresponding to each target The same is true, so Figure 1 The relevant processes are summarized in the figure, that is, the part from quantization → inverse super-prior transform → probability estimation is omitted. Of course, this is also the common representation method of the relevant processes in this field.

[0064] 4. Combining the overall discrete image features with the size information of the input image, the reconstructed image is obtained through the inverse transformation operation of the deep image decoder

[0065] In the embodiment of the present invention, the depth image decoder can be implemented through an existing network model, so it is not described in detail.

[0066] The above solution provided by the embodiment of the present invention has the following advantages:

[0067] (1) The introduction of a coding and decoding method with controllable entropy decoding complexity, combined with a highly scalable multi-complexity entropy coding scheme, enables the decoding of semantically structured code streams to support any pre-defined complexity requirements. It is suitable for intelligent analysis tasks in different application scenarios while introducing extremely low bit rates, thereby improving the versatility and flexibility of the semantically structured image coding method.

[0068] (2) For each target in the semantic structured image coding process, entropy coding reference dependencies of different complexities can be used for probabilistic modeling, which further enhances the flexibility and scalability of the framework and makes it more suitable for real application scenarios.

[0069] Example 2

[0070] The present invention also provides a coding and decoding system with controllable entropy decoding complexity, which is mainly implemented based on the method provided in the above embodiment, such as Figure 2 As shown, the system mainly includes:

[0071] an encoding unit configured to execute an encoding portion, the encoding portion comprising: performing target detection on an input image to obtain a target detection result; obtaining image features of the input image using a transform operation of a deep image encoder; performing a super-a priori transform and quantization on the image features to obtain discrete super-a priori features; encoding the size information of the input image, the discrete super-a priori features, and the target detection result, respectively, and then concatenating them to obtain a bitstream header; quantizing the image features to obtain overall discrete image features; grouping the overall discrete image features in combination with the target detection result to obtain discrete image features of each target; determining an autoregressive reference relationship corresponding to each target based on a predefined set of autoregressive reference relationships and a selected decoding complexity; determining a probability distribution of each target based on the autoregressive reference relationship of each target and the probability distribution of the overall discrete image features; performing entropy encoding on the discrete image features of the corresponding target to obtain a texture portion bitstream corresponding to each target; encoding an index of the autoregressive reference relationship of each target and combining it with the texture portion bitstream of the corresponding target to form a target-corresponding bitstream; and forming a semantically structured bitstream with the bitstream header information.

[0072] A decoding unit is used to execute the decoding part, which includes: entropy decoding the code stream header information in the semantically structured code stream to obtain the size information of the input image, the overall discrete super prior features and the target detection results; performing a super prior inverse transform on the overall discrete super prior features to obtain the probability distribution of the overall discrete image features, and the probability distribution of the overall discrete image features is used for entropy encoding of the discrete image features of each target in the encoding part and entropy decoding of the texture part code stream of each target; for each target, combining the probability distribution of the overall discrete image features to entropy decode the corresponding discrete image features from the target corresponding code stream, splicing the discrete image features corresponding to all targets according to the target detection results to obtain the overall discrete image features; then combining the overall discrete image features with the size information of the input image, and obtaining the reconstructed image through the inverse transform operation of the deep image decoder.

[0073] Those skilled in the art will clearly understand that for the convenience and brevity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above.

[0074] Example 3

[0075] The present invention also provides a processing device, such as Figure 3 As shown, it mainly includes: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the aforementioned embodiment.

[0076] Furthermore, the processing device further includes at least one input device and at least one output device; in the processing device, the processor, memory, input device, and output device are connected via a bus.

[0077] In the embodiment of the present invention, the specific types of the memory, input device, and output device are not limited; for example:

[0078] The input device can be a touch screen, image acquisition device, physical button or mouse;

[0079] The output device may be a display terminal;

[0080] The memory may be a random access memory (RAM) or a non-volatile memory, such as a disk memory.

[0081] Example 4

[0082] The present invention also provides a readable storage medium storing a computer program, which implements the method provided in the above embodiment when the computer program is executed by a processor.

[0083] In the embodiments of the present invention, the computer-readable storage medium may be provided in the aforementioned processing device, for example, as a memory in the processing device. Alternatively, the computer-readable storage medium may be a USB flash drive, a removable hard drive, a read-only memory (ROM), a magnetic disk, or an optical disk, among other media capable of storing program code.

[0084] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A coding and decoding method with controllable entropy decoding complexity, characterized in that: include: Encoding part: perform target detection on the input image and obtain target detection results; The image features of the input image are obtained by using a transformation operation of a deep image encoder, the image features are subjected to a super-prior transformation and quantization to obtain discrete super-prior features, the discrete super-prior features are subjected to a super-prior inverse transformation to obtain an overall probability distribution, and the size information of the input image, the discrete super-prior features, and the target detection results are respectively encoded and then spliced ​​to obtain codestream header information; the image features are quantized to obtain overall discrete image features, the overall discrete image features are grouped in combination with the target detection results to obtain discrete image features of each target, the autoregressive reference relationship corresponding to each target is determined based on a predefined set of autoregressive reference relationships and a selected decoding complexity, the probability distribution of each target is determined based on the autoregressive reference relationship of each target and the overall probability distribution, and the discrete image features of the corresponding target are entropy coded to obtain a texture portion codestream corresponding to each target, and the index of the autoregressive reference relationship of each target is encoded and then combined with the texture portion codestream of the corresponding target to form a target-corresponding codestream; all target-corresponding codestreams and the codestream header information constitute a semantically structured codestream; Decoding part: Entropy decoding is performed on the bitstream header information in the semantically structured bitstream to obtain the size information of the input image, the overall discrete super-prior features and the target detection results; the overall discrete super-prior features are inversely transformed to obtain the overall probability distribution; for each target, the corresponding discrete image features are entropy decoded from the target corresponding bitstream based on the overall probability distribution, and the discrete image features corresponding to all targets are spliced ​​according to the target detection results to obtain the overall discrete image features; then, the overall discrete image features are combined with the size information of the input image, and the reconstructed image is obtained through the inverse transformation operation of the deep image decoder.

2. The encoding and decoding method with controllable entropy decoding complexity according to claim 1, wherein: The performing target detection on the input image to obtain the target detection result includes: The target detection model is used to detect the input image and obtain the number of targets, the position and category of each target; wherein the position and category of the kth target are expressed as: k , b k , h k , w k ,cls k }, a k , b k , h k , w k ,cls k They represent the horizontal position of the upper left corner coordinate, the vertical position of the upper left corner coordinate, the height, the width and the category of the k-th target respectively.

3. The encoding and decoding method with controllable entropy decoding complexity according to claim 2, characterized in that: The encoding of the input image size information, discrete hyper-prior features, and target detection results, and then splicing them together to obtain the bitstream header information includes: The height H and width W of the input image, the discrete hyper-prior feature and target detection results are encoded separately, and then all the information obtained by encoding is spliced ​​in sequence to obtain the code stream header information; wherein, the discrete super prior features are When entropy coding is performed and target detection results are encoded separately, the number of targets, and the position and category of each target are encoded separately.

4. The encoding and decoding method with controllable entropy decoding complexity according to claim 1, wherein: Determining the autoregressive reference relationship corresponding to each target based on the predefined autoregressive reference relationship set and the selected decoding complexity includes: The predefined autoregressive reference relation set is denoted as R = {R1, R2, ..., R M }, R i Represents the i-th autoregressive reference relation. Different autoregressive reference relations correspond to different complexities. i is the index of the autoregressive reference relation, i = 1, 2, ..., M, M is the number of autoregressive reference relations. For the k-th target, the selected decoding complexity is recorded as T k , k = 1, 2, ..., K, K is the number of targets; For the kth target, combine the predefined autoregressive reference relation set R with the selected decoding complexity T k , determine the autoregressive reference relationship corresponding to the complexity as R k，i , where R k，i ∈R,k,i is the index of the autoregressive reference relation of the kth target, indicating the correspondence between the target and the autoregressive reference relation.

5. The encoding and decoding method with controllable entropy decoding complexity according to claim 1, wherein: Determining the probability distribution of each target based on the reference relationship of each target and the overall probability distribution, and performing entropy coding on the discrete image features of the corresponding target includes: For the kth target, the reference relation R of the kth target is combined with the autoregressive model k，i and the overall probability distribution Discrete image features for the kth target Perform probability modeling to obtain probability distribution Through the second entropy model, combined with probability distribution Discrete image features for the kth target Perform entropy coding.

6. The encoding and decoding method with controllable entropy decoding complexity according to claim 1, characterized in that: For each target, entropy decoding the corresponding discrete image features from the target corresponding bitstream in combination with the overall probability distribution includes: First, the index of the reference relationship is decoded from the entropy of the target corresponding code stream, and the autoregressive reference relationship is determined based on it; Combining the autoregressive reference relationship and the overall probability distribution, autoregressive entropy decoding is performed to obtain the discrete image features corresponding to each target.

7. A coding and decoding system with controllable entropy decoding complexity, characterized in that: The method according to any one of claims 1 to 6 is implemented, and the system comprises: The encoding unit is used to execute the encoding part, which includes: performing target detection on the input image to obtain the target detection result; using the transformation operation of the deep image encoder to obtain the image features of the input image, performing super-prior transformation and quantization on the image features to obtain discrete super-prior features, performing super-prior inverse transformation on the discrete super-prior features to obtain the overall probability distribution, and encoding the size information of the input image, the discrete super-prior features and the target detection results respectively, and then splicing them to obtain the code stream header information; quantizing the image features to obtain the overall discrete image features, combining the target detection results to obtain the overall discrete The image features are grouped to obtain discrete image features of each target. The autoregressive reference relationship corresponding to each target is determined based on a predefined set of autoregressive reference relationships and a selected decoding complexity. The probability distribution of each target is determined based on the autoregressive reference relationship of each target and the overall probability distribution. The discrete image features of the corresponding targets are entropy coded to obtain a texture portion bitstream corresponding to each target. The index of the autoregressive reference relationship of each target is encoded and then combined with the texture portion bitstream of the corresponding target to form a target corresponding bitstream. All target corresponding bitstreams and the bitstream header information constitute a semantically structured bitstream. A decoding unit is used to execute the decoding part, which includes: entropy decoding the code stream header information in the semantically structured code stream to obtain the size information of the input image, the overall discrete super prior features and the target detection results; performing a super prior inverse transform on the overall discrete super prior features to obtain the overall probability distribution; for each target, entropy decoding the corresponding discrete image features from the target corresponding code stream in combination with the overall probability distribution, splicing the discrete image features corresponding to all targets according to the target detection results to obtain the overall discrete image features; then combining the overall discrete image features with the size information of the input image to obtain the reconstructed image through the inverse transform operation of the deep image decoder.

8. A processing device, characterized in that include: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images