A deep learning-based image reconstruction method and system

Abstract

This invention relates to the field of image processing technology and discloses an image reconstruction method based on deep learning. The method includes: compressing an original image f; reconstructing the compressed original image f and determining an objective function based on Bayes' theorem; training a neural network model according to the objective function until the loss index of the neural network model meets a threshold condition; and reconstructing the image using the trained neural network model. This invention utilizes deep learning technology to provide higher compression rates and better image reconstruction quality than traditional methods.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an image reconstruction method and system based on deep learning. Background Technology

[0002] With the rapid development of information technology, data acquisition and processing have become key technologies in many fields. Traditional data acquisition methods follow the Nyquist sampling theorem, which requires the sampling frequency to be at least twice the highest frequency of the signal to ensure that the signal can be reconstructed without distortion. However, this method is inefficient and costly when processing high-dimensional data.

[0003] To address this challenge, compressed sensing theory was proposed in 2006, providing a new perspective for data acquisition and signal recovery. The core idea of ​​compressed sensing theory is to project high-dimensional data onto a low-dimensional space for acquisition, and then reconstruct the original signal from this low-dimensional data using specific algorithms. This method is based on a fundamental assumption: many natural signals can be compressed on some sparse basis, that is, most of the energy of the signal is concentrated on a few non-zero coefficients.

[0004] In the field of imaging, high-resolution detectors are usually very expensive, and the computational burden for acquiring and processing large amounts of data is often very heavy. Therefore, compressed sensing theory and its single-pixel camera technology have shown great advantages in this field. By reducing the use of high-resolution cameras, the system cost is reduced, while the image quality is guaranteed through efficient compressed acquisition and image reconstruction algorithms. In this process, the selection and optimization of image reconstruction methods are particularly important, as they are directly related to the quality and efficiency of the final imaging. Summary of the Invention

[0005] Therefore, it is necessary to propose a deep learning-based image reconstruction method to address the above problems.

[0006] A deep learning-based image reconstruction method, the method comprising:

[0007] Compress the original image f;

[0008] The compressed original image f is reconstructed, and the objective function is determined based on Bayes' theorem;

[0009] The neural network model is trained according to the objective function until the loss metric of the neural network model meets the threshold condition.

[0010] The image is reconstructed based on the trained neural network model.

[0011] In the above scheme, compressing the original image f specifically includes: generating a compression model based on the measurement signal g, the encoding matrix Φ, and the original image f.

[0012] p(Φ;f;g)=p(Φ)p(f)p(g)

[0013] Where p is the probability distribution function.

[0014] In the above scheme, after obtaining the compression model, the likelihood estimate of the compression model is obtained:

[0015]

[0016] in,‖·‖ 2 λ is a parameter, i is the i-th image patch, d is the number of image patches, and g i Let Φ be the i-th measurement signal, and Φ be the encoding matrix.

[0017] In the above scheme, after obtaining the likelihood estimate of the compression model, the measurement signal g is decomposed:

[0018]

[0019] Among them, D KL Let E be the KL divergence between the variational approximation distribution q(f, Φ|g) and the true posterior distribution p(f, Φ|g). q(f，Φ|g) Let f(f, Φ|g) be the expectation of the variational approximation distribution q(f, Φ|g).

[0020] In the above scheme, the reconstructing of the compressed original image f and the determination of the objective function based on Bayes' theorem specifically include:

[0021] The number of preset solutions is assumed to be the number of reconstructed images. It has a prior distribution:

[0022]

[0023] Where N(·;μ,σ 2 The mean is μ and the variance is σ. 2 The distribution is Gaussian, where I is the identity matrix and i is the i-th training image;

[0024] Constructing the variational form of the posterior of the reconstructed image And make it close to the posterior distribution

[0025] Assuming the reconstructed image The encoding matrix Φ is independently distributed:

[0026]

[0027] Estimate the posterior distribution in the form of parameters Determine the reconstructed image Variational posterior form:

[0028]

[0029] Among them, W R Here are the parameters of the deep Bayesian network R, g is the network input measurement signal, and μ is... i (g;W R ) and m 2 (g, W) R The mean and variance of the reconstructed image are respectively.

[0030] Obtain the target function:

[0031]

[0032] in, For divergence loss function, For the average error loss function, It is a constant.

[0033] In the above scheme, the construction of the variational form of the posterior of the reconstructed image... And make it close to the posterior distribution Also includes:

[0034] The first item The variational lower bound constituting the boundary likelihood p(g):

[0035]

[0036] in,

[0037] In the above scheme, the first likelihood term is:

[0038]

[0039]

[0040] Where const is a constant and λ is a parameter;

[0041] The second KL divergence term is analyzed as follows:

[0042]

[0043] Where σ is the variance, m i The mean of the reconstructed image.

[0044] In the above scheme, μ is 0 and σ is 0. 2 1×10-5 Let i be the i-th training image, i = 1, ..., d.

[0045] In the above scheme, the encoding matrix Φ satisfies the loss function: Where d is the number of image patches, I is the identity matrix, and Φ T This is the transpose of the encoding matrix.

[0046] This application also proposes a deep learning-based image reconstruction system, which includes: an objective function acquisition unit, a model training unit, and an image reconstruction unit;

[0047] The objective function acquisition unit is used to compress the original image f; reconstruct the compressed original image f; and determine the objective function based on Bayes' theorem.

[0048] The model training unit is used to train the neural network model according to the objective function until the loss index of the neural network model meets the threshold condition.

[0049] The image reconstruction unit is used to reconstruct the image based on the trained neural network model.

[0050] The embodiments of the present invention have the following beneficial effects: First, the original image f is compressed; the compressed original image f is reconstructed, and the objective function is determined according to Bayes' theorem; the neural network model is trained according to the objective function until the loss index of the neural network model meets the threshold condition; the image is reconstructed according to the trained neural network model; this method improves the image reconstruction quality by using a deep learning model and leveraging data advantages, and can provide a higher compression ratio and better image reconstruction quality than traditional methods, which is beneficial to improving image reconstruction efficiency. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] in:

[0053] Figure 1 This is a flowchart illustrating a deep learning-based image reconstruction method in one embodiment.

[0054] Figure 2 This is a schematic diagram of the target image compression measurement process;

[0055] Figure 3This is a schematic diagram of a deep Bayesian network.

[0056] Figure 4 For based on Figure 1 The reconstruction rendering. Detailed Implementation

[0057] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention; however, it will be apparent to those skilled in the art that the invention may be practiced without one or more of these details; in other instances, certain technical features well-known in the art have not been described in order to avoid confusion with the invention. It should be understood that the invention can be practiced in different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided to make the disclosure thorough and complete and to fully convey the scope of the invention to those skilled in the art.

[0059] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms, unless the context clearly indicates otherwise. The terms “comprising” and / or “including,” when used in this specification, identify the presence of said features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0060] In recent years, deep learning has developed rapidly in academia and industry. It has achieved significant improvements in recognition rates in many traditional signal representation and recognition tasks, demonstrating its ability to handle complex recognition tasks. This has attracted a large number of scholars to study its theory and applications, and many fields have begun to try to use deep learning to solve some problems in their respective fields.

[0061] In deep learning, convolutional neural networks (CNNs) and stacked denoising autoencoders possess excellent signal feature representation capabilities, making them suitable for compressed sensing. These networks accurately learn the structural features of real signals through a large number of training samples, rather than being limited by signal sparsity, thus improving signal reconstruction accuracy. Furthermore, with the support of parallel GPU computing hardware, the computation time of high-dimensional deep learning networks is guaranteed, allowing the compressed sensing reconstruction process to be converted into deep neural network computation, achieving real-time reconstruction. In the field of deep learning, CNNs, due to their powerful problem-solving capabilities, have been used extensively in almost all image-related research as a fundamental module for solving different types of problems. Autoencoders, on the other hand, exhibit excellent performance in both high-dimensional and low-dimensional data processing.

[0062] To facilitate understanding, the relevant terms used in this application will be introduced below.

[0063] (1) Deep variational Bayes refers to a technique that combines deep learning and variational Bayesian inference. In deep learning, we often use large neural network models to handle complex tasks, but these models are prone to overfitting on limited datasets and their generalization ability may be weak. To address this issue, Bayesian Deep Learning was proposed, which attempts to introduce model uncertainty into neural networks to improve the robustness and generalization ability of the model.

[0064] (2) Prior distribution: In Bayesian statistics and Bayesian deep learning, the prior distribution represents our beliefs or assumptions about a parameter or a set of parameters before we see any data. The prior distribution is the probability distribution of the possible values ​​of the parameter, reflecting our "prior" knowledge of these possible values;

[0065] (3) Posterior distribution, which describes the probability distribution of unknown parameters under given observation data. The posterior distribution is the result of Bayesian inference, which is calculated by combining the prior distribution and the likelihood function.

[0066] (4) Encoding Matrix: In machine learning and data science, the encoding matrix can be viewed as a K-row, L-column matrix consisting of codes for all categories, where K represents the number of sample categories in the multi-class task and L represents the code length. This means that the multi-class task is divided into L binary classification tasks. The codes in the encoding matrix are generally referred to as "codewords." During training, each binary classification task divides the dataset according to the positive and negative class labels of the corresponding columns of the encoding matrix and trains a classifier, ultimately producing L binary classifiers.

[0067] To fully understand the present invention, a detailed structure will be presented in the following description in order to illustrate the technical solution proposed by the present invention; optional embodiments of the present invention are described in detail below, however, in addition to these detailed descriptions, the present invention may have other embodiments.

[0068] like Figure 1 As shown, in one embodiment, a deep learning-based image reconstruction method is provided, which includes steps S101 to S104, detailed below:

[0069] S101. Compress the original image f;

[0070] In some embodiments, compressing the original image f specifically includes:

[0071] A compression model is generated based on the measurement signal g, the encoding matrix Φ, and the original image f:

[0072] p(Φ;f;g)=p(Φ)p(f)p(g)

[0073] Where p is the probability distribution function.

[0074] Specifically, the compression model described above was generated within a Bayesian framework.

[0075] Preferably, after obtaining the compressed model, the likelihood estimate of the compressed model is obtained:

[0076]

[0077] in,‖·‖ 2 λ is a parameter, i is the i-th image patch, d is the number of image patches, and g i Let Φ be the i-th measurement signal, and Φ be the encoding matrix.

[0078] In some embodiments, after obtaining the likelihood estimate of the compression model, the measurement signal g is decomposed:

[0079]

[0080] Among them, D KL Let E be the KL divergence between the variational approximation distribution q(f, Φ|g) and the true posterior distribution p(f, Φ|g). q(f，Φ|g) Let f(f, Φ|g) be the expectation of the variational approximation distribution q(f, Φ|g).

[0081] like Figure 2 The above-mentioned target image compression measurement process is shown.

[0082] S102. Reconstruct the compressed original image f and determine the objective function based on Bayes' theorem;

[0083] The objective function is key to the optimization problem. It defines how to evaluate the performance of a neural network model. In image reconstruction tasks, the objective function may involve pixel-level error, structural similarity, or other perceptual quality metrics. Once the objective function is determined, it can ensure that the neural network model optimizes in the right direction during training, thereby producing high-quality reconstructed images.

[0084] In some embodiments, the compressed original image f is reconstructed, and the objective function is determined according to Bayes' theorem, specifically including:

[0085] (1) The number of preset solutions is assumed to be the number of reconstructed images. It has a prior distribution:

[0086]

[0087] Where N(·;μ,σ 2 The mean is μ and the variance is σ. 2 The distribution is Gaussian, where I is the identity matrix and i is the i-th training image;

[0088] This step is to enable the reconstruction of the image. The solution should be as close as possible to the original image f, so the number of solutions is limited. Assume... It has a prior distribution, specifically, μ is 0 and σ is 0. 2 1×10 -5 Let i be the i-th training image, i = 1, ..., d.

[0089] (2) Constructing the variational form of the posterior of the reconstructed image And make it close to the posterior distribution

[0090] (3) Assuming the image is reconstructed The encoding matrix Φ is independently distributed:

[0091]

[0092] (4) Estimate the posterior distribution in the form of parameters. Determine the reconstructed image Variational posterior form:

[0093]

[0094] Among them, W R Here are the parameters of the deep Bayesian network R, g is the network input measurement signal, and μ is... i (g;W R ) and m 2 (g, W) R ) represent the mean and variance of the reconstructed image, respectively;

[0095] (5) Obtain the objective function:

[0096]

[0097] in, For divergence loss function, For the average error loss function, It is a constant.

[0098] 10. Preferably, the above encoding matrix Φ satisfies the loss function: Where d is the number of image patches, I is the identity matrix, and Φ T This is the transpose of the encoding matrix.

[0099] In some embodiments, a variational form of the reconstructed image posterior is constructed. And make it close to the posterior distribution Also includes:

[0100] The first item The variational lower bound constituting the boundary likelihood p(g):

[0101]

[0102] in,

[0103] Preferably, the first likelihood term is:

[0104]

[0105] Where const is a constant and λ is a parameter;

[0106] The second KL divergence term is analyzed as follows:

[0107]

[0108] Where σ is the variance and mi is the mean of the reconstructed image.

[0109] S103. Train the neural network model according to the objective function until the loss index of the neural network model meets the threshold condition.

[0110] Training a neural network model is an iterative process in which the model is tuned based on the performance of the objective function on the training data. In image reconstruction tasks, training data may include the original image and its corresponding compressed version. By training a neural network model, it is possible to learn the complex mappings required to recover the original image from compressed data, generalize to unseen data, and produce high-quality reconstructed images.

[0111] like Figure 3As shown, the aforementioned neural network model is a deep Bayesian network, comprising a sampling subnetwork, an initial reconstruction subnetwork, and a deep reconstruction subnetwork. The sampling subnetwork consists of a 3×3 convolutional layer, with the training weights being the optimized encoding aperture. The initial reconstruction subnetwork also consists of a 3×3 convolutional layer, performing an upsampling operation from two-dimensional measurements to a three-dimensional data cube. The deep reconstruction subnetwork consists of a 3×3 convolutional layer, a feature space information supplementation module, and a 1×1 convolutional layer, effectively removing noise during the reconstruction process to obtain a high-quality reconstructed image.

[0112] Preferably, the above threshold conditions can be manually set and adjusted according to different determining factors such as data and usage scenarios.

[0113] S104. Reconstruct the image based on the trained neural network model.

[0114] Once the neural network model is trained and meets performance thresholds, it can be used for practical image reconstruction tasks. This involves receiving new compressed images as input and outputting the corresponding reconstructed images. Since the neural network model has been trained to optimize a specific objective function, it should be able to produce high-quality, realistic reconstructed images. Furthermore, due to the powerful capabilities of deep learning models, this method can potentially handle various types of images and compression algorithms.

[0115] like Figure 4As shown: Images labeled 0 and 1 are the original images; images labeled a and g are images processed by an instant learning network, with PSNR / SSIM values ​​of 28.09 / 0.8864 and 2939 / 0.8759 respectively; images labeled b and k are images processed by an optimization-inspired interpretable deep network, with PSNR / SSIM values ​​of 29.34 / 0.9155 and 31.17 / 0.9121 respectively; images labeled c and l are images processed using a combination of average pooling and max pooling. Images processed by neural network architectures using two pooling strategies (pooling) have PSNR / SSIM values ​​of 29.19 / 0.9054 and 31.28 / 0.9096, respectively. Images labeled d and m are processed by a COAST neural network, with PSNR / SSIM values ​​of 28.30 / 0.8929 and 30.04 / 0.8848, respectively. Images labeled e and n are processed by a neural network architecture based on the classic U-Net architecture but with added dynamic and guided elements, with PSNR / SSIM values ​​of 29.38 / 0.9214 and 31.92 / 0.9189, respectively. Images labeled f and o are processed by a hybrid architecture model based on Transformer, with PSNR / SSIM values ​​of 29.43 / 0.9134 and 31.98 / 0.9184, respectively. Images labeled g and p are processed by a neural network architecture based on the reference article: Dynamic... Images processed by the Path-Controllable Deep Unfolding Network for Compressive Sensing show PSNR / SSIM values ​​of 29.05 / 0.9040 and 30.88 / 0.8977, respectively. Images labeled h and q are processed based on the reference article "Deep memory-augmented proximal unrolling network for compressive sensing," with PSNR / SSIM values ​​of 28.98 / 0.9032 and 30.97 / 0.8990, respectively. Images labeled i and r represent the reconstruction results of this invention, with PSNR / SSIM values ​​of 29.78 / 0.9257 and 32.20 / 0.9204, respectively. It is evident that compared to other deep learning reconstruction algorithms, the reconstruction results of this invention achieve optimal PSNR and SSIM. Furthermore, magnified views of the reconstruction results show that the reconstruction of details and textures by this invention is better restored, without blocky effects or blurring.

[0116] In summary, the deep learning-based image reconstruction method proposed in this application establishes the probability distribution of the reconstructed image, obtains the maximum likelihood estimate of the compressed observation model, constructs a variational form of the posterior distribution to approximate the posterior distribution of the reconstructed image, designs a deep reconstruction network to learn the parameters of this variational form, and finally sets the loss function of the deep network as the variational lower bound of the boundary likelihood, thereby obtaining the reconstructed target image through network training.

[0117] This application also proposes a deep learning-based image reconstruction system, which includes: an objective function acquisition unit, a model training unit, and an image reconstruction unit;

[0118] The objective function acquisition unit is used to compress the original image f; reconstruct the compressed original image f; and determine the objective function based on Bayes' theorem.

[0119] The model training unit is used to train the neural network model according to the objective function until the loss metric of the neural network model meets the threshold condition.

[0120] The image reconstruction unit is used to reconstruct images based on a trained neural network model.

[0121] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments can be accomplished by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.

[0122] Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0123] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0124] The embodiments described above are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application's patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. The embodiments disclosed above are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made according to the claims of this invention are still within the scope of this invention.

Claims

1. A deep learning-based image reconstruction method, characterized in that, The method includes: The original image f is compressed, specifically including: A compression model is generated based on the measurement signal g, the encoding matrix Φ, and the original image f: Where p is the probability distribution function; The original image f is projected into a low-dimensional space using the encoding matrix Φ to obtain the compressed signal, which is reflected in the likelihood estimation formula of the compression model as follows: in, λ is a positive real parameter related to the observation noise level, i is the i-th image patch, d is the number of image patches, and g i Let Φ be the i-th measurement signal, and Φ be the encoding matrix. The compressed original image f is reconstructed based on a deep learning network, and the objective function is determined according to Bayes' theorem. The deep learning network is a deep Bayesian network, which includes a sampling sub-network, an initial reconstruction sub-network, and a deep reconstruction sub-network. The sampling sub-network consists of a 3x3 convolutional layer, the initial reconstruction sub-network also consists of a 3x3 convolutional layer, and the deep reconstruction sub-network consists of a 3x3 convolutional layer, a feature space information supplementation module, and a 1x1 convolutional layer. The neural network model is trained according to the objective function until the loss metric of the neural network model meets the threshold condition. The image is reconstructed based on the trained neural network model.

2. The image reconstruction method based on deep learning according to claim 1, characterized in that, After obtaining the likelihood estimate of the compression model, the measurement signal g is decomposed: in, D KL For variational approximation distribution and the true posterior distribution Between KL divergence, For variational approximation distribution Expectations It is a variational lower bound function.

3. The image reconstruction method based on deep learning according to claim 1, characterized in that, The process of reconstructing the compressed original image f and determining the objective function based on Bayes' theorem specifically includes: The number of preset solutions is assumed to be the number of reconstructed images. It has a prior distribution: in, N (·; μ , σ 2 (The mean is) μ The variance is σ 2 Gaussian distribution, I It is the identity matrix. i For the first i One training image; Constructing the variational form of the posterior of the reconstructed image And make it close to the posterior distribution ; Assuming the reconstructed image Independently distributed with respect to the encoding matrix Φ: ； Estimate the posterior distribution in the form of parameters Determine the reconstructed image Variational posterior form: in, W R For the parameters of the deep Bayesian network R, g Input measurement signals to the network, and These are the mean and variance of the reconstructed image, respectively. Obtain the target function: in, For divergence loss function, For the average error loss function, It is a constant.

4. The image reconstruction method based on deep learning according to claim 3, characterized in that, The variational form of constructing the posterior of the reconstructed image. And make it close to the posterior distribution Also includes: The first item Constructing boundary likelihood p ( g Variational lower bound of ) in, .

5. The image reconstruction method based on deep learning according to claim 4, characterized in that, The first likelihood term is: Where const is a constant and λ is a parameter; The second KL divergence term is analyzed as follows: Where σ is the variance, m i The mean of the reconstructed image.

6. The image reconstruction method based on deep learning according to claim 4, characterized in that, The μ 0, σ 2 1×10 -5 , i For the first i One training image, i =1,… d .

7. The image reconstruction method based on deep learning according to claim 6, characterized in that, The encoding matrix Φ satisfies the loss function: Where d is the number of image patches, I is the identity matrix, and Φ T This is the transpose of the encoding matrix.

8. A deep learning-based image reconstruction system, characterized in that, The system includes: an objective function acquisition unit, a model training unit, and an image reconstruction unit; The objective function acquisition unit is used to compress the original image f, specifically including: A compression model is generated based on the measurement signal g, the encoding matrix Φ, and the original image f: Where p is the probability distribution function; The original image f is projected into a low-dimensional space using the encoding matrix Φ to obtain the compressed signal, which is reflected in the likelihood estimation formula of the compression model as follows: in, λ is a positive real parameter related to the observation noise level, i is the i-th image patch, d is the number of image patches, and g i Let Φ be the i-th measurement signal, and Φ be the encoding matrix. The compressed original image f is reconstructed based on a deep learning network, and the objective function is determined according to Bayes' theorem. The deep learning network is a deep Bayesian network, which includes a sampling sub-network, an initial reconstruction sub-network, and a deep reconstruction sub-network. The sampling sub-network consists of a 3x3 convolutional layer, the initial reconstruction sub-network also consists of a 3x3 convolutional layer, and the deep reconstruction sub-network consists of a 3x3 convolutional layer, a feature space information supplementation module, and a 1x1 convolutional layer. The model training unit is used to train the neural network model according to the objective function until the loss index of the neural network model meets the threshold condition. The image reconstruction unit is used to reconstruct the image based on the trained neural network model.

Images