CN112801906B - Loop Iterative Image Denoising Method Based on Recurrent Neural Network - Google Patents

Abstract

The invention relates to a cyclic iterative image denoising method based on a cyclic neural network, which comprises the following steps of S1, obtaining paired original noise images and noiseless images and preprocessing the images to obtain paired image blocks of the noise images and the noiseless images for training; s2, constructing a cyclic iterative image denoising network based on a cyclic neural network, and training by using paired image blocks of a noise image and a noiseless image; and S3, inputting the original noise image to be detected into the trained denoising network to obtain a denoising image. The invention removes noise more cleanly and keeps more image details in a circular iteration mode, thereby effectively reconstructing a de-noised image.

Description

Loop Iterative Image Denoising Method Based on Recurrent Neural Network

technical field

The invention relates to the technical field of image and video processing, in particular to a cyclic iterative image denoising method based on a cyclic neural network.

Background technique

With the rapid development of the Internet and multimedia technology, images have become an indispensable part of the process of human information exchange and information delivery. Image has research value in communication, social and medical fields, and has practical significance for the development of information storage and information interaction technology in modern society. However, image content will inevitably degrade, such as image degradation caused by camera parameter settings, image degradation caused by ambient brightness, image quality degradation caused by image compression and decompression techniques, etc. The degraded image will seriously affect the beauty of the image and even cause the image information cannot be extracted effectively. Once the outline of the object is not clear in the image content, the foreground and background of the image will not be effectively segmented, and even more seriously, the content of the image will not be able to be distinguished. Therefore, it is necessary to process the degraded image. Image denoising is one of the essential techniques to reconstruct the degraded image to get closer to the noise-free image content. As a low-level visual task, its calculation results can directly affect advanced computer vision tasks such as image segmentation, image classification, and target recognition.

The goal of image denoising is to reconstruct the image content in the noisy image, so that the denoised image has more image detail information. The image denoising task has a long research history. Among the image denoising methods that have been proposed, they can be roughly divided into traditional methods and methods based on deep learning. Traditional methods use filters such as median filter and Gaussian filter to process noisy images, but due to the limitation of computing resources and the need to manually extract image priors, the processing efficiency is low. This method needs some processing and optimization before it can be applied to real life. The method based on deep learning utilizes the automatic feature extraction ability of the convolutional neural network, and can use the prior information extracted by traditional methods to help the convolutional neural network to extract the features of the image. Therefore, methods based on deep learning have been widely studied by researchers in recent years.

In recent years, with the improvement of computer computing power, methods based on deep learning have developed rapidly. Image denoising methods based on deep learning have been continuously proposed, and have more advanced denoising performance than traditional methods. However, many current image denoising methods still have some problems. For example, the denoising result is too smooth, the texture loss is serious, etc. If only one denoising operation is performed on the noisy image to obtain the denoising result, it is easy to cause the denoising result to be too smooth and lose image details, and the result cannot be reversed. If multiple iterative denoising processes are performed on the noisy image, a part of the noise of the noisy image can be removed each time until the denoising result obtained by iteration can remove more image noise and reconstruct more image textures. And if the noise distribution in the noisy image is reasonably estimated before the denoising operation, then the estimated noise distribution information is simultaneously input into the denoising network. The noise amplitude and noise position of the noise image can be estimated and positioned more accurately, so that the denoising operation can be performed on the noise image better, and a denoising result with better performance can be obtained.

Contents of the invention

In view of this, the object of the present invention is to provide a cyclic iterative image denoising method based on a cyclic neural network, which can remove noise more cleanly and retain more image details through cyclic iterations, thereby effectively reconstructing the denoised image. image.

To achieve the above object, the present invention adopts the following technical solutions:

A cyclic iterative image denoising method based on recurrent neural network, comprising the following steps:

Step S1: obtain paired original noise image and noise-free image and preprocess, obtain the paired image blocks of noise image and noise-free image for training;

Step S2: construct the cyclic iterative image denoising network based on the cyclic neural network, and use the paired image block training of the noise image and the noise-free image;

Step S3: Input the original noise image to be tested into the trained denoising network to obtain a denoising image.

The step S1 is specifically:

Step S11: cutting the paired original noise image and noise-free image into blocks at the same position to obtain image blocks of pairs of noise images and noise-free images;

Step S12: performing the same random flip and rotation on each pair of obtained image blocks, and enhancing the data to obtain a pair of image blocks of a noise image and a noise-free image for training.

Further, the cyclic iterative image denoising network based on the cyclic neural network includes a noise estimation sub-network and a denoising sub-network.

Further, the noise estimation sub-network consists of an input convolutional layer, m series-connected ResNet residual blocks, a GRU module and an output convolutional layer. Among them, the input convolution layer is composed of a convolution kernel with a convolution kernel of 3×3 and a step size of 1, and the output convolution layer is composed of a convolution kernel with a convolution kernel of 1×1 and a step size of 1; the GRU module The input of is the output of the GRU module of the previous iteration and the output of the ResNet residual block of the current iteration number i, which is the feature z obtained by channel splicing.

Further, the specific calculation formula of the GRU module is as follows:

a=conv(z),

b=conv(z),

c = conv_(z),

d=conv(z),

作为输出卷积层的输入，

为第i+1次迭代的GRU模块输入的一部分。Among them, conv(·) contains a convolution layer with a convolution kernel of 3 and a step size of 1 and a sigmoid activation function, conv_(·) contains a convolution kernel with a convolution kernel of 3 and a step size of 1 and a convolution and tanh activation function,

As input to the output convolutional layer,

Part of the input to the GRU module for iteration i+1.

Further, the denoising sub-network includes two loop operations, each loop operation is composed of the same encoder, residual module and decoder:

The encoder consists of an input convolutional layer and two downsampling layers; the input convolutional layer contains a convolutional layer with a convolution kernel of 5×5 and a stride of 1 and a GRU module, and the downsampling layer contains a convolutional layer with convolution A convolutional layer with a product kernel of 5×5 and a step size of 2, an activation function, and a GRU module; the input of the GRU module is the output of the corresponding feature size of the previous layer and the output of the GRU module of the corresponding feature size of the previous iteration through the channel The features obtained by splicing are calculated in the same way as the GRU module of the noise estimation sub-network; the encoder part divides the features of the network into three different scales, from large scale to small scale are F ₁ , F ₂ and F ₃ ;

The residual module consists of n series-connected ResNet residual blocks, its input is the feature F ₃ obtained by the encoder part, and the output feature is F ^c ;

The decoder consists of two upsampling layers and an output convolutional layer. The operation of each upsampling layer includes a nearest neighbor interpolation operation, a convolutional layer with a kernel size of 3×3 and a stride of 1, and a ReLU activation function, the output convolutional layer is a convolutional layer with a convolution kernel size of 1×1 and a step size of 1; the input of the first upsampling layer is the feature obtained by F ₃ and F ^c through the channel splicing operation, The output feature is f ₃ ; the input of the second upsampling layer is the feature obtained by F ₂ and f ₃ through the channel splicing operation, and the output feature is f ₂ ; the input of the output convolution layer is F ₁ and f ₂ through the channel The features obtained by the stitching operation are output as a denoised image.

Further, the denoising network includes t loop iterations, and a part of each GRU module input of the first loop iteration is a feature whose corresponding feature size is 0. From the second loop iteration, each GRU A part of the module input is the GRU feature of the last cycle iteration corresponding to the feature size; and the input of the noise estimation subnetwork in the first cycle iteration is the noise image N _ori , and the output is the noise estimation image E ₁ of the first cycle iteration ; The input of the denoising sub-network in the first loop iteration is the noise estimation image E ₁ of the first loop iteration and the noise image N _ori after the channel splicing of the dual-channel image, and the output is the denoising of the first loop iteration Image De ₁ . The noise estimation sub-network starts from the ith (i>1) loop iteration, the input of the i-th loop iteration is the denoised image De i-1 of the _i-1 loop iteration, and the output is the i-th loop iteration’s Noise estimated image E _i . The denoising sub-network starts from the second loop iteration, and the input of the i-th loop iteration is the noise estimation image E _i of the i-th loop iteration and the denoising image De _{i-1 of the i-1} loop iteration through channel splicing The final two-channel image is output as the denoising image De _i of the i-th stage. The final denoised image is the output De _t of the t-th loop iteration.

Further, the training of the cyclic iterative image denoising network model based on the cyclic neural network is specifically:

(1): Randomly divide the image blocks of the paired noise image and the noise-free image into multiple batches, and each batch contains N image blocks;

(2): taking the batch as a unit, input the noise images of the corresponding N image blocks in the batch into the denoising network described in step S2, and obtain corresponding N denoising images;

(3): According to the target loss function of the cyclic iterative image denoising network based on the cyclic neural network, use the back propagation method to calculate the gradient of each parameter in the network, and use the stochastic gradient descent method to update the parameters of the network;

(4): Repeat the above (2)-(3) in batches until the value of the target loss function of the cyclic iterative image denoising network based on the cyclic neural network tends to be stable, save the network parameters, and complete the training process of the network .

Further, the target loss function of the cyclic iterative image denoising network based on the cyclic neural network is calculated as follows:

Loss＝λ ₁ L _stg1 +λ ₂ L _stg2 +...+λ _k L _stgk

In the target loss function, λ ₁ , λ ₂ , ..., λ _k are the weights of the loss for each loop iteration, and L _stg1 is the loss function for the first loop iteration, and the specific calculation is as follows:

Among them, N _gt1 represents the reference image of the noise estimation sub-network in the first cycle iteration, that is, the difference image between the original noise image and the noise-free image, f( ) represents the noise estimation sub-network, and N _ori represents the The input image, w _f represents the model parameters of the noise estimation sub-network, I _gt represents the reference image of the noise image, g( ) represents the denoising sub-network, concat( ) represents the channel concatenation operation, E ₁ represents the first loop iteration In the output image of the noise estimation sub-network, w _g represents the model parameters of the denoising sub-network, ||·|| ₁ represents the L ₁ distance, ||·|| ₂ represents the L ₂ distance;

In the target loss function, L _stg2 and L _stgk are the loss functions of the second and k-th loop iterations, starting from the i-th (i>1) loop iteration, the loss function L _stgi of the i-th loop iteration is specifically calculated as follows:

Among them, N _gti represents the reference image of the noise estimation sub-network in the i-th loop iteration, that is, the difference image between the output denoised image and the noise-free image of the i-th loop iteration, f( ) denotes the noise estimation sub-network, De _i-1 represents the output denoised image of the i-th cycle iteration, which is also the input image of the i-th cycle iteration, w _f represents the model parameters of the noise estimation sub-network, I _gt represents the reference image of the noise image, and g( ) represents denoising sub-network, concat(·) represents the channel splicing operation, E _i represents the output image of the noise estimation sub-network in the i-th loop iteration, w _g represents the model parameters of the denoising sub-network, ||·|| ₁ represents L ₁ distance, || · || ₂ means L ₂ distance.

Compared with the prior art, the present invention has the following beneficial effects:

The present invention uses a multi-scale encoder and a residual module to effectively extract the features of a noisy image and reconstruct a denoised image through a decoder. By means of loop iteration, the noise is removed more cleanly while retaining more image details, so as to effectively reconstruct the denoised image.

Description of drawings

Fig. 1 is a schematic flow sheet of the method of the present invention;

2 is a schematic diagram of the overall structure of the network training in step S2 of the embodiment of the present invention (k=2);

3 is a schematic diagram of the network structure of the noise estimation sub-network during the first iteration of the embodiment of the present invention;

Fig. 4 is a schematic diagram of the network structure of the denoising sub-network in the first iteration of the embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Please refer to Fig. 1, the present invention provides a kind of cyclic iteration image denoising method based on cyclic neural network, comprises the following steps:

Step S1: obtain paired original noise image and noise-free image and preprocess, obtain the paired image blocks of noise image and noise-free image for training;

Step S2: construct the cyclic iterative image denoising network based on the cyclic neural network, and use the paired image block training of the noise image and the noise-free image;

Step S3: Input the original noise image to be tested into the trained denoising network to obtain a denoising image.

In this embodiment, step S1 is specifically:

Step S11: cutting the paired original noise image and noise-free image into blocks at the same position to obtain image blocks of pairs of noise images and noise-free images;

Step S12: performing the same random flip and rotation on each pair of obtained image blocks, and enhancing the data to obtain a pair of image blocks of a noise image and a noise-free image for training.

In this embodiment, the cyclic iterative image denoising network based on the cyclic neural network includes a noise estimation subnetwork and a denoising subnetwork.

Preferably, the noise estimation sub-network consists of an input convolutional layer, m connected ResNet residual blocks, a GRU module and an output convolutional layer. Among them, the input convolution layer is composed of a convolution kernel with a convolution kernel of 3×3 and a step size of 1, and the output convolution layer is composed of a convolution kernel with a convolution kernel of 1×1 and a step size of 1; the GRU module The input of is the output of the GRU module of the previous iteration and the output of the ResNet residual block of the current iteration number i, which is the feature z obtained by channel splicing.

The specific calculation formula of the GRU module is as follows:

a=conv(z),

b=comv(z),

c = conv_(z),

d=conv(z),

作为输出卷积层的输入，

为第i+1次迭代的GRU模块输入的一部分。Among them, conv(·) contains a convolution layer with a convolution kernel of 3 and a step size of 1 and a sigmoid activation function, conv_(·) contains a convolution kernel with a convolution kernel of 3 and a step size of 1 and a convolution and tanh activation function,

As input to the output convolutional layer,

Part of the input to the GRU module for iteration i+1.

Preferably, the denoising sub-network contains two loop operations, each loop operation consists of the same encoder, residual module and decoder:

The encoder consists of an input convolutional layer and two downsampling layers; the input convolutional layer contains a convolutional layer with a convolution kernel of 5×5 and a stride of 1 and a GRU module, and the downsampling layer contains a convolutional layer with convolution A convolutional layer with a product kernel of 5×5 and a step size of 2, an activation function, and a GRU module; the input of the GRU module is the output of the corresponding feature size of the previous layer and the output of the GRU module of the corresponding feature size of the previous iteration through the channel The features obtained by splicing are calculated in the same way as the GRU module of the noise estimation sub-network; the encoder part divides the features of the network into three different scales, from large scale to small scale are F ₁ , F ₂ and F ₃ ;

The residual module consists of n series-connected ResNet residual blocks, its input is the feature F ₃ obtained by the encoder part, and the output feature is F ^c ;

The decoder consists of two upsampling layers and an output convolutional layer. The operation of each upsampling layer includes a nearest neighbor interpolation operation, a convolutional layer with a kernel size of 3×3 and a stride of 1, and a ReLU activation function, the output convolutional layer is a convolutional layer with a convolution kernel size of 1×1 and a step size of 1; the input of the first upsampling layer is the feature obtained by F ₃ and F ^c through the channel splicing operation, The output feature is f ₃ ; the input of the second upsampling layer is the feature obtained by F ₂ and f ₃ through the channel splicing operation, and the output feature is f ₂ ; the input of the output convolution layer is F ₁ and f ₂ through the channel The features obtained by the stitching operation are output as a denoised image.

The denoising network contains t loop iterations. A part of the input of each GRU module in the first loop iteration is a feature whose corresponding feature size is 0. From the second loop iteration, a part of the input of each GRU module is The GRU feature of the last loop iteration corresponding to the feature size; and the input of the noise estimation subnetwork in the first loop iteration is the noise image N _ori , and the output is the noise estimation image E ₁ of the first loop iteration; the denoising subnetwork The input of the first loop iteration is the noise estimation image E ₁ of the first loop iteration and the noise image N _ori after channel splicing, and the output is the denoised image De ₁ of the first loop iteration. The noise estimation sub-network starts from the ith (i>1) loop iteration, the input of the i-th loop iteration is the denoised image De i-1 of the _i-1 loop iteration, and the output is the i-th loop iteration’s Noise estimated image E _i . The denoising sub-network starts from the second loop iteration, and the input of the i-th loop iteration is the noise estimation image E _i of the i-th loop iteration and the denoising image De _{i-1 of the i-1} loop iteration through channel splicing The final two-channel image is output as the denoising image De _i of the i-th stage. The final denoised image is the output De _t of the t-th loop iteration.

Preferably, the cyclic iteration image denoising network model training based on the cyclic neural network is specifically:

(1): Randomly divide the image blocks of the paired noise image and the noise-free image into multiple batches, and each batch contains N image blocks;

(2): taking the batch as a unit, input the noise images of the corresponding N image blocks in the batch into the denoising network described in step S2, and obtain corresponding N denoising images;

(3): According to the target loss function of the cyclic iterative image denoising network based on the cyclic neural network, use the back propagation method to calculate the gradient of each parameter in the network, and use the stochastic gradient descent method to update the parameters of the network;

(4): Repeat the above (2)-(3) in batches until the value of the target loss function of the cyclic iterative image denoising network based on the cyclic neural network tends to be stable, save the network parameters, and complete the training process of the network .

Preferably, in this embodiment, the target loss function of the cyclic iterative image denoising network based on the cyclic neural network is calculated as follows:

Loss＝λ ₁ L _stg1 +λ ₂ L _stg2 +...+λ _k L _stgk

In the target loss function, λ ₁ , λ ₂ , ..., λ _k are the weights of the loss for each loop iteration, and L _stg1 is the loss function for the first loop iteration, and the specific calculation is as follows:

Among them, N _gt1 represents the reference image of the noise estimation sub-network in the first cycle iteration, that is, the difference image between the original noise image and the noise-free image, f( ) represents the noise estimation sub-network, and N _ori represents the The input image, w _f represents the model parameters of the noise estimation sub-network, I _gt represents the reference image of the noise image, g( ) represents the denoising sub-network, concat( ) represents the channel concatenation operation, E ₁ represents the first loop iteration In the output image of the noise estimation sub-network, w _g represents the model parameters of the denoising sub-network, ||·|| ₁ represents the L ₁ distance, ||·|| ₂ represents the L ₂ distance;

In the target loss function, L _stg2 and L _stgk are the loss functions of the second and k-th loop iterations, starting from the i-th (i>1) loop iteration, the loss function L _stgi of the i-th loop iteration is specifically calculated as follows:

Among them, N _gti represents the reference image of the noise estimation sub-network in the i-th loop iteration, that is, the difference image between the output denoised image and the noise-free image of the i-th loop iteration, f( ) denotes the noise estimation sub-network, De _i-1 represents the output denoised image of the i-th cycle iteration, which is also the input image of the i-th cycle iteration, w _f represents the model parameters of the noise estimation sub-network, I _gt represents the reference image of the noise image, and g( ) represents denoising sub-network, concat(·) represents the channel splicing operation, E _i represents the output image of the noise estimation sub-network in the i-th loop iteration, w _g represents the model parameters of the denoising sub-network, ||·|| ₁ represents L ₁ distance, || · || ₂ means L ₂ distance.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

The above descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope of the present invention.

Claims (6)

1. A cyclic iteration image denoising method based on a cyclic neural network is characterized by comprising the following steps:

s1, acquiring and preprocessing paired original noise images and noiseless images to obtain paired image blocks of the noise images and the noiseless images for training;

s2, constructing a cyclic iterative image denoising network based on a cyclic neural network, and training by using paired image blocks of a noise image and a noiseless image;

s3, inputting the original noise image to be detected into the trained denoising network to obtain a denoising image;

the recurrent iterative image denoising network based on the recurrent neural network comprises a noise estimation sub-network and a denoising sub-network;

the denoising sub-network comprises t times of cyclic operations, and each cyclic operation comprises the same encoder, a residual error module and a decoder:

the noise estimation sub-network consists of an input convolutional layer, m series-connected ResNet residual blocks, a GRU module and an output convolutional layer; wherein, the input convolution layer is composed of convolution kernels with convolution kernel of 3 multiplied by 3 and step length of 1; the output convolution layer is composed of convolution kernels with convolution kernel of 1 multiplied by 1 and step length of 1; the input of the GRU module is a characteristic z obtained by splicing the output of the GRU module of the previous iteration and the output of the ResNet residual block of the current iteration number i through a channel;

the encoder consists of an input convolutional layer and two down-sampling layers; the input convolutional layer comprises a convolutional layer with a convolutional kernel of 5 multiplied by 5 and a step size of 1 and a GRU module, and the downsampling layer comprises a convolutional layer with a convolutional kernel of 5 multiplied by 5 and a step size of 2, an activation function and a GRU module; the input of the GRU module is the characteristics obtained by channel splicing of the output of the corresponding characteristic dimension of the previous layer and the output of the GRU module of the corresponding characteristic dimension of the previous iteration, and the calculation mode is the same as that of the GRU module of the noise estimation subnetwork; the encoder part divides the characteristics of the network into 3 different scales, which are respectively F from large scale to small scale ₁ 、F ₂ And F ₃ ；

The residual module is composed of n series ResNet residual blocks, and the input of the residual module is the characteristic F obtained by the encoder part ₃ The output is characterized by F ^c ；

The decoder consists of two up-sampling layers and an output convolution layer, wherein the operation of each up-sampling layer comprises one-time nearest neighbor interpolation operation, a convolution layer with the convolution kernel size of 3 multiplied by 3 and the step length of 1 and a ReLU activation function, and the output convolution layer is a convolution layer with the convolution kernel size of 1 multiplied by 1 and the step length of 1; the input to the first up-sampling layer is F ₃ And F ^c The output characteristic is f ₃ (ii) a The input of the second up-sampling layer is F ₂ And f ₃ The output characteristic is f ₂ (ii) a The input of the output convolutional layer is F ₁ And f ₂ And outputting the characteristics obtained by the channel splicing operation as a de-noised image.

2. The method for denoising cyclic iterative images based on the recurrent neural network as claimed in claim 1, wherein the step S1 specifically comprises:

step S11: the original noise image and the noiseless image which are paired are cut into blocks at the same position, and a plurality of groups of image blocks of the noise image and the noiseless image which are paired are obtained;

step S12: and carrying out the same random inversion and rotation on each group of paired image blocks, and enhancing the data to obtain the image blocks of paired noisy images and noiseless images for training.

3. The method of claim 1, wherein the GRU module has a specific calculation formula as follows:

a＝conv(z)，

b＝conv(z)，

c＝conv_(z)，

d＝conv(z)，

wherein conv (·) comprises a convolution layer with convolution kernel of 3 and step size of 1 and a sigmoid activation function, conv _ (·) comprises a convolution layer with convolution kernel of 3 and step size of 1 and a tanh activation function,

as an input to the output convolutional layer,

is part of the GRU module input for the (i + 1) th iteration.

4. The recurrent neural network-based recurrent iterative image denoising method of claim 1, wherein the denoising network comprises t recurrent iterations, and a portion of each GRU module input of a first recurrent iteration is a value of a corresponding feature sizeA feature of 0, starting from the second iteration of the loop, a portion of each GRU module input is the GRU feature of the last iteration of the loop corresponding to the feature size; and the input of the noise estimation subnetwork at the first iteration of the loop is a noise image N _ori Output as a noise estimate image E for the first iteration of the loop ₁ (ii) a The input of the denoising subnetwork in the first iteration of the loop is a noise estimation image E of the first iteration of the loop ₁ And a noise image N _ori Outputting the two-channel image after channel splicing as a De-noised image De of the first loop iteration ₁ (ii) a The noise estimation sub-network starts from the ith (i is more than 1) loop iteration, and the input of the ith loop iteration is the denoised image De of the (i-1) loop iteration _i-1 And outputting a noise estimation image E of the ith loop iteration _i (ii) a The denoising subnetwork starts from the second loop iteration, and the input of the ith loop iteration is a noise estimation image E of the ith loop _i And the De-noised image De of the i-1 th loop iteration _i-1 Outputting the two-channel image spliced by the channels as a De-noised image De of the ith stage _i (ii) a The final denoised image is the output De of the t-th iteration of the loop _t 。

5. The method for denoising cyclic iterative images based on the cyclic neural network as claimed in claim 1, wherein the training of the cyclic iterative image denoising network model based on the cyclic neural network is specifically:

(1) Randomly dividing image blocks of a pair of a noisy image and a noiseless image into a plurality of batches, each batch comprising N image blocks;

(2): inputting the noise images of the corresponding N image blocks in the batch into the denoising network in the step S2 by taking the batch as a unit to obtain corresponding N denoising images;

(3): calculating the gradient of each parameter in the network by using a back propagation method according to a target loss function of a cyclic iterative image denoising network based on a cyclic neural network, and updating the parameters of the network by using a random gradient descent method;

(4): and (3) repeating the steps (2) to (3) by taking batches as units until the target loss function value of the recurrent iterative image denoising network based on the recurrent neural network tends to be stable, storing the network parameters and finishing the training process of the network.

6. The method of claim 5, wherein the objective loss function of the recurrent neural network-based image denoising network is calculated as follows:

Loss＝λ ₁ L _stg1 +λ ₂ L _stg2 +...+λ _k L _stgk

in said objective loss function, λ ₁ ，λ ₂ ，...，λ _k Weight lost for each iteration of the loop, L _stg1 For the loss function of the first loop iteration, the specific calculation is as follows:

wherein, N _gt1 Representing the reference image of the noise estimation sub-network in the first iteration of the loop, i.e. the difference image of the original noisy image and the clean image, f (-) representing the noise estimation sub-network, N _ori Representing the input image of the first cycle, w _f Model parameters representing a noise estimation sub-network, I _gt A reference image representing a noisy image, g (-) representing a denoising subnetwork, concat (-) representing a channel splicing operation, E ₁ Output image, w, representing a noise estimation sub-network in a first iteration of the loop _g Model parameters representing a de-noising subnetwork, | ·| non-woven phosphor ₁ Represents L ₁ Distance, | · | luminance ₂ Represents L ₂ A distance;

l in the objective loss function _stg2 ，L _stgk For the loss function of the second and kth loop iteration, starting from the ith (i > 1) loop iteration, the loss function L of the ith loop iteration _stgi The specific calculation is as follows:

wherein N is _gti A reference image representing the noise estimation subnetwork in the ith iteration of the loop, i.e. the difference image of the output denoised image and the noiseless image of the ith iteration of the loop, f (-) represents the noise estimation subnetwork, de _i-1 The output denoised image representing the i-1 st iteration of the loop, which is also the input image for the i-th iteration of the loop, w _f Model parameters representing noise estimation sub-network, I _gt A reference image representing a noisy image, g (-) representing a denoising subnetwork, concat (-) representing a channel splicing operation, E _i Representing the output image of the noise estimation sub-network in the ith iteration of the loop, w _g Model parameters representing a de-noising subnetwork, | ·| non-woven phosphor ₁ Represents L ₁ Distance, | · | luminance ₂ Represents L ₂ Distance.

Images