Image processing method and device based on sparse representation, terminal and storage medium
By using a pre-trained deep learning model and a fixed overcomplete dictionary, the problem of high computational cost in sparse representation image processing is solved, achieving more efficient image processing.
Patent Information
- Application Number
- CN202110444628.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-04-24
- Publication Date
- 2025-12-05
- Estimated Expiration
- 2041-04-24
AI Technical Summary
Existing image processing techniques based on sparse representations involve large computational demands, requiring iterative optimization to solve for sparse representation coefficients, resulting in high computational complexity.
A pre-trained deep learning model is used to determine sparse representation coefficients, and a fixed overcomplete dictionary is used to extract target features. The target image is obtained through convolution processing, avoiding iterative optimization.
It reduces the computational cost of sparse representation image processing, improves processing efficiency, and can more effectively describe image information.
Smart Images
Figure CN115240017B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of image processing, in particular to an image processing method and device based on sparse expression, a terminal and a storage medium. BACKGROUND
[0002] Sparse is one of the most basic characteristics of signals, which helps to describe more sufficient information with the least information. Research on sparse expression has achieved many research results in various fields such as signal processing, system identification and image processing. In the traditional image processing technology based on sparse expression, an over-complete dictionary and sparse expression coefficients need to be determined. In the prior art, the sparse expression coefficients of an image need to be solved by iterative optimization, which requires a large amount of calculation.
[0003] Therefore, the prior art still needs to be improved and enhanced. SUMMARY
[0004] In view of the above defects of the prior art, the present application provides an image processing method and device based on sparse expression, a terminal and a storage medium, aiming to solve the problem of large amount of calculation in the prior art image processing technology based on sparse expression.
[0005] In a first aspect, the present application provides an image processing method based on sparse expression, comprising:
[0006] providing an image to be processed, determining the sparse expression coefficients corresponding to the image to be processed by a trained deep learning model;
[0007] obtaining the target features corresponding to the image to be processed according to the sparse expression coefficients and a preset over-complete dictionary;
[0008] obtaining the target image corresponding to the image to be processed according to the target features.
[0009] The image processing method based on sparse expression, wherein the determination of the sparse expression coefficients corresponding to the image to be processed by the trained deep learning model comprises:
[0010] cutting the image to be processed into a plurality of image blocks to be processed, and extracting the initial features corresponding to each of the image blocks to be processed by the trained deep learning model;
[0011] performing first convolution processing on the initial features corresponding to each of the image blocks to be processed by the trained deep learning model to obtain the sparse expression coefficients corresponding to each of the image blocks to be processed.
[0012] The image processing method based on sparse expression, wherein the obtaining of the target features corresponding to the image to be processed according to the sparse expression coefficients and the preset over-complete dictionary comprises:
[0013] According to the sparse expression coefficient corresponding to each of the to-be-processed image blocks and the preset overcomplete dictionary, a target feature corresponding to each of the to-be-processed image blocks is obtained.
[0014] The image processing method based on sparse expression, wherein the target image corresponding to the to-be-processed image is obtained according to the target feature, comprises:
[0015] The target feature corresponding to each of the to-be-processed image blocks is subjected to second convolution processing by the trained deep learning model, so as to obtain a target image block corresponding to each of the to-be-processed image blocks.
[0016] The target image is obtained by performing image reconstruction on the target image block corresponding to each of the to-be-processed image blocks.
[0017] The image processing method based on sparse expression, wherein the target feature corresponding to each of the to-be-processed image blocks is subjected to second convolution processing by the trained deep learning model, so as to obtain a target image block corresponding to each of the to-be-processed image blocks, comprises:
[0018] When the resolution requirement of the target image is higher than that of the to-be-processed image, the target feature corresponding to each of the to-be-processed image blocks is subjected to second convolution processing and then up-sampling processing, so as to obtain a target image block corresponding to each of the to-be-processed image blocks.
[0019] The image processing method based on sparse expression, wherein the trained deep learning model is obtained according to a plurality of groups of training data, and each group of training data comprises a sample to-be-processed image and a sample target image corresponding to the sample to-be-processed image.
[0020] The image processing method based on sparse expression, wherein before the to-be-processed image is input into the trained deep learning model, the method further comprises:
[0021] The preset overcomplete dictionary is generated according to a preset kernel function.
[0022] The second aspect of the application provides an image processing device based on sparse expression, comprising:
[0023] A sparse expression unit is configured to determine, by a trained deep learning model, a sparse expression coefficient corresponding to a to-be-processed image provided by a user.
[0024] A target feature acquisition unit is configured to obtain a target feature of the to-be-processed image according to the sparse expression coefficient and a preset overcomplete dictionary.
[0025] An object image acquisition unit is configured to acquire an object image corresponding to the to-be-processed image according to the object feature.
[0026] In a third aspect, the present application provides a terminal, comprising: a processor, and a storage medium connected with the processor, the storage medium being adapted to store a plurality of instructions, and the processor being adapted to invoke the instructions in the storage medium to execute the steps of the image processing method based on sparse representation.
[0027] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores one or more programs, and the one or more programs are executable by one or more processors to implement the steps of the image processing method based on sparse representation.
[0028] Beneficial effects: compared with the prior art, the present application provides an image processing method based on sparse representation, device, terminal and storage medium, the image processing method based on sparse representation provided by the present application predicts the sparse representation coefficient of the to-be-processed image through a pre-trained deep learning model, and extracts the object feature of the to-be-processed image according to a fixed over-complete dictionary to output an object image, without solving the sparse representation coefficient through an iterative optimization method, thereby reducing the calculation amount in the image processing based on sparse representation. BRIEF DESCRIPTION OF DRAWINGS
[0029] Figure 1 A flow chart of an embodiment of the image processing method based on sparse representation provided by the present application;
[0030] Figure 2 A structural schematic diagram of a feature extraction model in an embodiment of the image processing method based on sparse representation provided by the present application;
[0031] Figure 3 An effect schematic diagram of an embodiment of the image processing method based on sparse representation provided by the present application Figure 1 ;
[0032] Figure 4 An effect schematic diagram of an embodiment of the image processing method based on sparse representation provided by the present application Figure 2 ;
[0033] Figure 5 A structural schematic diagram of an embodiment of the image processing device based on sparse representation provided by the present application;
[0034] Figure 6 A structural schematic diagram of an embodiment of the terminal provided by the present application. DETAILED DESCRIPTION
[0035] For the purposes of the present application, the technical solutions and effects are more clear and explicit, the following will be further described in detail with reference to the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application, and are not used to limit the present application.
[0036] The image processing method based on sparse expression provided by the present application can be applied in a terminal, which can be but is not limited to various personal computers, notebook computers, mobile phones, tablet computers, vehicle-mounted computers and portable wearable devices. After obtaining a to-be-processed image, the terminal can process the to-be-processed image to obtain a target image by using the image processing method based on sparse expression provided by the present application.
[0037] Embodiment one
[0038] As shown in Figure 1 The image processing method based on sparse expression provided by the present application comprises the following steps:
[0039] S100, providing a to-be-processed image, and determining a sparse expression coefficient corresponding to the to-be-processed image by using a trained deep learning model.
[0040] The image processing method based on sparse expression provided by the present embodiment can be applied in various image processing tasks, such as image super-resolution and denoising. Sparse is one of the most basic characteristics of a signal, which helps to describe more sufficient information with the least information. In sparse expression, the most basic concept is an over-complete dictionary D∈R n×K , which is composed of a large number of redundant sources (K>n). Assuming that a signal y can be expressed as a sparse linear combination on an over-complete dictionary, the process can be described as: y=Dα0, wherein the sparse expression coefficient α0 has only a small number of non-zero values.
[0041] In the traditional image processing technology based on sparse expression, the sparse expression coefficient α0 needs to be solved by using an iterative optimization method. The solving process of the sparse expression modeling leads to high computational complexity and large amount of calculation, and it is difficult to truly land. In the present embodiment, a deep learning model is used to obtain sparse expression.
[0042] Specifically, the determination of the sparse expression coefficient corresponding to the to-be-processed image by using the trained deep learning model comprises:
[0043] S110, dividing the to-be-processed image into a plurality of to-be-processed image blocks, and extracting initial features of each to-be-processed image block by using the trained deep learning model;
[0044] S120, respectively, the trained deep learning model is used to perform first convolution processing on the initial features corresponding to each of the to-be-processed image blocks, so as to obtain sparse expression coefficients corresponding to each of the to-be-processed image blocks.
[0045] The trained deep learning model can be a CNN (Convolutional Neural Network) model. The initial features of the trained deep learning model can be pixel values of each pixel point of the to-be-processed image, or can be obtained by a feature extraction module. The architecture of the feature extraction module can be consistent with that of an existing deep learning neural network, for example, including one or more convolution layers. The initial features of the to-be-processed image are obtained by performing convolution processing on the pixel values of each pixel point of the to-be-processed image through the one or more convolution layers. The parameters of the feature extraction module are determined in the process of training the deep learning model.
[0046] In a possible implementation, in order to further improve the image processing quality, the to-be-processed image is divided into a plurality of to-be-processed image blocks for processing. After receiving the to-be-processed image, the to-be-processed image is divided into a plurality of to-be-processed image blocks. For each to-be-processed image block, the initial features can be obtained by the foregoing method of extracting target features of the to-be-processed image, that is, using the pixel values of each pixel point in each to-be-processed image block as initial features or performing convolution on the pixel values of each pixel point to obtain initial features.
[0047] After obtaining the initial features of each to-be-processed image block, the trained deep learning model is used to perform first convolution processing on the initial features corresponding to each of the to-be-processed image blocks, so as to obtain sparse expression coefficients corresponding to each of the to-be-processed image blocks. Specifically, the trained deep learning model can be provided with a sparse expression module. The sparse expression module includes one or more convolution layers, and the initial features of the to-be-processed image block are processed by the one or more convolution layers. The specific parameters of the sparse expression module are determined in the training process of the trained deep learning model, so that the sparse expression module has the ability to obtain corresponding sparse expression coefficients according to the input to-be-processed image block.
[0048] After obtaining the sparse expression coefficients corresponding to the to-be-processed image, the image processing method based on sparse expression provided by the embodiment further includes the following steps:
[0049] S200, obtaining target features corresponding to the to-be-processed image according to the sparse expression coefficients and a preset over-complete dictionary.
[0050] Specifically, the target feature corresponding to the to-be-processed image includes target features corresponding to the to-be-processed image blocks, and the target feature corresponding to the to-be-processed image is obtained according to the sparse expression coefficient and the preset overcomplete dictionary.
[0051] The target feature corresponding to each to-be-processed image block is obtained according to the sparse expression coefficient corresponding to the to-be-processed image block and the preset overcomplete dictionary.
[0052] After the sparse expression coefficient corresponding to the to-be-processed image block is obtained, the target feature is obtained through the aforementioned sparse expression formula y=Da0, that is, the sparse expression coefficient and the preset overcomplete dictionary are multiplied to obtain the target feature of the to-be-processed image block. It is not difficult to see that in this embodiment, the sparse expression coefficient corresponding to the to-be-processed image block is obtained through the trained deep learning model, without the need to determine the sparse expression coefficient in a traditional iterative optimization manner, so that the calculation amount of implementing sparse expression can be reduced, and meanwhile, sparse expression can realize that less information is used to describe more sufficient information. After the target feature of the to-be-processed image is obtained through the sparse expression manner, the target feature of the to-be-processed image is processed, so that the data calculation amount can be further reduced.
[0053] Please refer to Figure 1 The image processing method based on sparse expression provided in this embodiment further includes the steps of:
[0054] S300, obtaining a target image corresponding to the to-be-processed image according to the target feature.
[0055] Specifically, the target image corresponding to the to-be-processed image is obtained according to the target feature, including:
[0056] S310, performing second convolution processing on the target features corresponding to each to-be-processed image block through the trained deep learning model to obtain a target image block corresponding to each to-be-processed image block.
[0057] S320, performing image reconstruction on the target image blocks corresponding to each to-be-processed image block to obtain the target image.
[0058] Specifically, the trained deep learning model can be provided with an output module, and the output module includes at least one convolution layer. The target features of the to-be-processed image blocks are input into the output module, and the convolution layers in the output module perform convolution processing (that is, the second convolution processing) on the target features corresponding to each to-be-processed image block respectively to output a target image block corresponding to each to-be-processed image block. After the target image blocks corresponding to each to-be-processed image block are obtained, image reconstruction is performed on each target image block to obtain the target image.
[0059] For image processing tasks such as image denoising, the resolution of the image is unchanged, that is, the size is unchanged, while for image super-resolution tasks, the resolution of the image needs to be increased, therefore, for image super-resolution tasks, the step of upsampling also needs to be added, that is, the second convolution processing is respectively performed on the target features corresponding to each of the image blocks to be processed to obtain the target image blocks corresponding to each of the image blocks to be processed, comprising:
[0060] When the resolution requirement of the target image is higher than that of the image to be processed, after the second convolution processing is performed on the target features corresponding to each of the image blocks to be processed, upsampling processing is performed to obtain the target image blocks corresponding to each of the image blocks to be processed.
[0061] As shown in Figure 2 In a possible implementation, if the resolution requirement of the target image is higher than that of the image to be processed, that is, the image processing task performed by using the image processing method based on sparse representation provided in the embodiment is an image super-resolution task, the output module includes an upsample layer, and the upsample layer can perform upsampling by using an existing deep learning upsample implementation to improve the resolution.
[0062] Specifically, the image processing method based on sparse representation provided in the embodiment can realize end-to-end training, that is, the trained deep learning model is obtained by training a plurality of sets of training data, and each set of training data includes a sample to-be-processed image and a sample target image corresponding to the sample to-be-processed image. The sample target image corresponding to the sample to-be-processed image in the training data used to train the deep learning model can be selected according to a specific image processing task. For example, when the image processing method based on sparse representation provided in the embodiment is used to implement an image super-resolution task, the sample target image corresponding to the sample to-be-processed image should be an image with a higher resolution than the sample to-be-processed image; when the image processing method based on sparse representation provided in the embodiment is used to implement an image denoising task, the sample target image corresponding to the sample to-be-processed image should be an image with a consistent resolution as the sample to-be-processed image but without noise, and so on. The parameters of the trained deep learning model, such as the parameters of the feature extraction module, the sparse representation module and the output module, are determined in the training process. The preset overcomplete dictionary can be determined in the training process of the trained deep learning model. In one possible implementation, an initial overcomplete dictionary can be generated first, the initial overcomplete dictionary is embedded into the deep learning model after being generated, and the training of the deep learning model is started. In the training process of the deep learning model, the elements in the overcomplete dictionary are also updated. The overcomplete dictionary after being trained is used as the preset overcomplete dictionary to obtain the target image corresponding to the to-be-processed image. However, the number of elements in the overcomplete dictionary is large, and the way of generating the preset overcomplete dictionary through training will cause too many variable parameters of the model and low training efficiency. In the embodiment, the preset overcomplete dictionary can be directly generated, and the elements in the preset overcomplete dictionary are fixed, that is, the elements in the overcomplete dictionary remain unchanged in the process of training the deep learning model, so that other modules in the trained deep learning model can match the preset overcomplete dictionary to achieve the purpose of obtaining the target image corresponding to the to-be-processed image. That is, before the trained deep learning model determines the sparse representation coefficient corresponding to the to-be-processed image, the image processing method based on sparse representation provided in the embodiment further includes the steps of:
[0063] generating the preset overcomplete dictionary according to a preset kernel function.
[0064] The preset kernel function can be a different Gaussian kernel function, a same Gaussian kernel function, a DOG kernel function, and the like. The way of generating an overcomplete dictionary according to a preset kernel function is a prior art, and will not be described here.
[0065] The overcomplete dictionary generated according to the preset kernel function is directly used as the preset overcomplete dictionary, and the overcomplete dictionary used in the training process remains unchanged, so that the number of variable parameters in the deep learning model can be reduced. In an actual experiment taking an image super-resolution task as an example, an L1 loss function is used for supervision, two data sets, DIV2K and Flickr2K, are used for model training, the overcomplete dictionary remains unchanged during the training process, and only less than 150k parameters are required to obtain a super-resolution effect comparable to or even better than the current super-resolution algorithm model, FALSR algorithm model (Xiaomi open source image super-resolution algorithm model), CARN (Cascading Residual Network, cascading residual network) model, and the calculation amount is less (about 25% of CARN).
[0066] As Figures 3-4 shown, Figure 3 and Figure 4 respectively, Figure 3 the bicubic processing result, the processing result of the image processing method based on sparse expression provided in the embodiment, and the corresponding sample target image (GroudTruth, GT) from left to right in the comparison, Figure 4 the input to-be-processed image, the processing result of the image processing method based on sparse expression provided in the embodiment, and the corresponding sample target image (GroudTruth, GT) from left to right in the comparison, Figure 3 and Figure 4 It can be seen that the method provided in the embodiment can well restore the texture and details of the image.
[0067] In summary, the image processing method based on sparse expression provided in the embodiment predicts the sparse expression coefficient of the to-be-processed image through the pre-trained deep learning model, and extracts the target feature of the to-be-processed image according to the fixed overcomplete dictionary in the trained deep learning model to output the target image. The calculation amount in the image processing based on sparse expression is reduced without solving the sparse expression coefficient by iterative optimization.
[0068] It should be understood that, although each step in the flowchart shown in the drawings of the present application specification is shown in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of the steps in the present application is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps of the present application can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or sub-steps or stages of other steps.
[0069] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of each method. Wherein, any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM) and the like.
[0070] Embodiment two
[0071] Based on the above-mentioned embodiments, the present application also provides an image processing device based on sparse expression, and a structure principle diagram of the device can be as shown in the figure Figure 5 The device comprises:
[0072] A processing unit is configured to provide a to-be-processed image, and determine a sparse expression coefficient corresponding to the provided to-be-processed image through a trained deep learning model, as described in embodiment one.
[0073] The target feature acquisition unit is used to acquire the target features of the image to be processed based on the sparse representation coefficients and a preset overcomplete dictionary, as specifically described in Embodiment 1;
[0074] A target image acquisition unit is used to acquire the target image corresponding to the image to be processed based on the target features, as described in Embodiment 1. Embodiment 3
[0075] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 6 As shown. The terminal includes a processor 10 and a memory 20, the memory 20 storing a computer program. When the processor 10 executes the computer program, it can perform at least the following steps:
[0076] Provide an image to be processed, and determine the sparse representation coefficients corresponding to the image to be processed through a trained deep learning model;
[0077] The target features corresponding to the image to be processed are obtained based on the sparse representation coefficients and the preset overcomplete dictionary;
[0078] The target image corresponding to the image to be processed is obtained based on the target features.
[0079] Wherein, determining the sparse representation coefficients corresponding to the image to be processed through a trained deep learning model includes:
[0080] The image to be processed is divided into several image blocks, and the initial features corresponding to each image block are extracted by the trained deep learning model.
[0081] The trained deep learning model performs a first convolution on the initial features corresponding to each of the image blocks to be processed, thereby obtaining the sparse representation coefficients corresponding to each of the image blocks to be processed.
[0082] The step of obtaining the target features corresponding to the image to be processed based on the sparse representation coefficients and the preset overcomplete dictionary includes:
[0083] The target features corresponding to each image block to be processed are obtained based on the sparse representation coefficients corresponding to each image block to be processed and the preset overcomplete dictionary.
[0084] Wherein, obtaining the target image corresponding to the image to be processed based on the target features includes:
[0085] The trained deep learning model performs a second convolution process on the target features corresponding to each of the image blocks to be processed, thereby obtaining the target image blocks corresponding to each of the image blocks to be processed.
[0086] performing image reconstruction on the target image blocks corresponding to each of the to-be-processed image blocks to obtain the target image.
[0087] The second convolutional processing on the target features corresponding to each of the to-be-processed image blocks comprises:
[0088] When the resolution requirement of the target image is higher than that of the to-be-processed image, after the second convolutional processing on the target features corresponding to each of the to-be-processed image blocks, upsampling processing is performed to obtain the target image blocks corresponding to each of the to-be-processed image blocks.
[0089] The trained deep learning model is obtained according to a plurality of groups of training data, and each group of training data comprises a sample to-be-processed image and a sample target image corresponding to the sample to-be-processed image.
[0090] Before the step of determining the sparse representation coefficients corresponding to the to-be-processed image by using the trained deep learning model, the method further comprises:
[0091] generating the preset over-complete dictionary according to a preset kernel function.
[0092] Embodiment Four
[0093] The application further provides a computer readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the image processing method based on sparse representation.
[0094] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.
Claims
1. A method of image processing based on sparse representation, characterized in that, The method comprises the following steps: providing a to-be-processed image, determining a sparse expression coefficient corresponding to the to-be-processed image by using a trained deep learning model, wherein the trained deep learning model comprises an output module; obtaining a target feature corresponding to the to-be-processed image according to the sparse expression coefficient and a preset overcomplete dictionary; obtaining a target image corresponding to the to-be-processed image according to the target feature; the step of obtaining the target image corresponding to the to-be-processed image according to the target feature comprises the following steps: inputting the target feature corresponding to the to-be-processed image into the output module, so that a convolution layer in the output module performs convolution processing on the target feature corresponding to the to-be-processed image, and outputs a target image block corresponding to the to-be-processed image; performing image reconstruction on the target image block to obtain the target image.
2. The sparse representation based image processing method of claim 1, wherein, the step of determining the sparse expression coefficient corresponding to the to-be-processed image by using the trained deep learning model comprises the following steps: dividing the to-be-processed image into a plurality of to-be-processed image blocks, and extracting an initial feature corresponding to each to-be-processed image block by using the trained deep learning model; performing first convolution processing on the initial feature corresponding to each to-be-processed image block by using the trained deep learning model, and obtaining a sparse expression coefficient corresponding to each to-be-processed image block.
3. The sparse representation based image processing method of claim 2, wherein, the step of obtaining the target feature corresponding to the to-be-processed image according to the sparse expression coefficient and the preset overcomplete dictionary comprises the following step: obtaining a target feature corresponding to each to-be-processed image block according to the sparse expression coefficient corresponding to each to-be-processed image block and the preset overcomplete dictionary.
4. The sparse representation based image processing method of claim 3, wherein, the step of obtaining the target image corresponding to the to-be-processed image according to the target feature comprises the following steps: performing second convolution processing on the target feature corresponding to each to-be-processed image block by using the trained deep learning model, and obtaining a target image block corresponding to each to-be-processed image block; performing image reconstruction on the target image block corresponding to each to-be-processed image block to obtain the target image.
5. The sparse representation based image processing method of claim 4, wherein, the step of performing second convolution processing on the target feature corresponding to each to-be-processed image block by using the trained deep learning model to obtain the target image block corresponding to each to-be-processed image block comprises the following step: when the resolution requirement of the target image is higher than that of the to-be-processed image, performing second convolution processing on the target feature corresponding to each to-be-processed image block and then performing up-sampling processing to obtain the target image block corresponding to each to-be-processed image block.
6. The sparse representation based image processing method of claim 5, wherein, The trained deep learning model is trained according to a plurality of groups of training data, and each group of training data comprises a sample to-be-processed image and a sample target image corresponding to the sample to-be-processed image.
7. The sparse representation based image processing method according to any one of claims 1-6, characterized in that, Before the step of determining the sparse expression coefficient corresponding to the to-be-processed image by using the trained deep learning model, the method further comprises the following step: generating the preset overcomplete dictionary according to a preset kernel function.
8. An image processing apparatus based on sparse representation, characterized by, The method comprises the following steps: a sparse expression unit is configured to determine a sparse expression coefficient corresponding to a to-be-processed image provided by using a trained deep learning model, wherein the trained deep learning model comprises an output module; The target feature acquisition unit is configured to acquire a target feature of the image to be processed according to the sparse expression coefficient and a preset over-complete dictionary. The target image acquisition unit is configured to acquire a target image corresponding to the image to be processed according to the target feature, and the acquiring of the target image corresponding to the image to be processed according to the target feature comprises: inputting the target feature corresponding to the image to be processed into the output module, so that a convolution layer in the output module performs convolution processing on the target feature corresponding to the image to be processed, and outputs a target image block corresponding to the image to be processed; performing image reconstruction on the target image block to obtain the target image.
9. A terminal, characterized by comprising: The terminal comprises a processor and a storage medium in communication connection with the processor, the storage medium is adapted to store a plurality of instructions, and the processor is adapted to call the instructions in the storage medium to execute the steps of the image processing method based on sparse expression according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the image processing method based on sparse expression according to any one of claims 1-7.