Image encryption and decryption method based on dynamic compressed sensing

Abstract

The invention relates to an image encryption and decryption method based on dynamic compressed sensing, and belongs to the technical field of image information security. The method comprises the following steps: uniformly segmenting an image with a preset size, inputting image sub-blocks into a compression network, and after two-dimensional Logistic chaotic mapping, carrying out sliding scrambling encryption and XOR diffusion encryption to obtain a final ciphertext image; and after the ciphertext image is subjected to inverse diffusion and inverse scrambling, inputting the ciphertext image into a deep reconstruction network to obtain reconstructed sub-block images, and splicing the sub-block images to realize image restoration. According to the method, image compression, encryption and decryption can be realized, the requirement on image sparsity during compression and encryption is reduced, and the problem that high-quality reconstruction is difficult to realize at a low sampling rate is solved by utilizing the self-learning capability of the deep convolutional neural network.

Description

Technical Field

[0001] The present invention belongs to the technical field of image information security, and in particular relates to an image encryption and decryption method based on dynamic compressed sensing. Background Art

[0002] As an important information carrier, images are widely used in various fields. The security issues in the storage and transmission of sensitive images have arisen, which has attracted widespread attention from researchers. Generally, images have high pixel redundancy to represent rich information, which increases the resource overhead of storage and network transmission. Therefore, compressing images and then encrypting and decrypting them has become a crucial technology.

[0003] At present, image encryption algorithms based on dynamic compressed sensing theory usually use chaotic mapping to generate compression matrices, compress sparse images and encrypt them to obtain observation signals, which are then further encrypted to obtain the final ciphertext image. This method has problems such as the constructed random compression matrix does not have dynamic adaptability, the image to be processed requires sparse transformation, and high-quality reconstruction cannot be achieved at low sampling rates.

[0004] In recent years, artificial intelligence technology has developed rapidly, and convolutional neural networks (CNNs) applicable to the image field have attracted widespread attention from researchers. The present invention combines block compressed sensing and convolutional neural networks to construct a dynamic adaptive compression network; designs a reconstruction network based on convolutional neural networks; thereby getting rid of the encryption algorithm's requirements for signal sparsity and sampling rate, while enhancing the image reconstruction quality. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] The technical problem to be solved by the present invention is how to provide an image encryption and decryption method based on dynamic compressed sensing to solve the problems that the constructed random compression matrix does not have dynamic adaptability, the image to be processed requires sparse transformation, and high-quality reconstruction cannot be achieved at low sampling rates.

[0007] (II) Technical solution

[0008] In order to solve the above technical problems, the present invention proposes an image encryption and decryption method based on dynamic compressed sensing, which comprises the following steps:

[0009] Step S1, input an original image and divide it into multiple sub-images, each sub-image has a size of N×N;

[0010] Step S2: using a dynamic compression network to perform convolution compression and initialization reconstruction on the sub-image to obtain a preliminary encrypted image;

[0011] Step S3, using the SHA256 algorithm to calculate the 64-byte Hash value of the sub-image, dividing the Hash value with a step length of 2 bytes, using the two numbers of the first 4 bytes as the common factor α, and dividing the remaining 60 bytes into β1 and β2; using α to perform an XOR operation on β1 and β2 to obtain two 2-byte numbers, converting them into decimal and dividing them by 255 to obtain two initial keys as the initial values ​​x0 and y0 of the two-dimensional Logistic chaotic system;

[0012] Step S4: The two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 to generate two chaotic sequences x n and n , from the chaotic sequence x n 32×2 numbers are cut out and enlarged and rounded to the interval [0,32]. The first 32 numbers are used for row sliding scrambling of the preliminary encrypted image, and the last 32 numbers are used for column sliding scrambling of the preliminary encrypted image.

[0013] Then from the chaotic sequence y n Take 32×32×2 numbers, enlarge them to the interval [0, 255] and divide them into two parts as the random key for diffusion encryption;

[0014] After sliding scrambling and diffusion encryption, the final encrypted ciphertext image is obtained;

[0015] Step S5, using two random sequences to decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain a decrypted image;

[0016] Step S6: reconstruct the decrypted image through a nonlinear deep reconstruction network to restore the sub-image;

[0017] Step S7: The sub-images are spliced ​​in the order of cutting to restore the original image.

[0018] (III) Beneficial effects

[0019] The present invention proposes an image encryption and decryption method based on dynamic compressed sensing. The beneficial effects of the present invention are as follows:

[0020] (1) Reduced requirements for image sparsity during compression and encryption: The present invention uses a convolutional layer that removes activation functions and biases to achieve image compression, thereby ignoring whether the image is sparse;

[0021] (2) Utilizing the self-learning ability of deep convolutional neural networks to improve the problem of difficulty in achieving high-quality reconstruction at low sampling rates: The present invention obtains an image reconstruction network model that implements nonlinear operations by training a deep convolutional neural network. This model can obtain more feature information from low-sampling rate images, thereby achieving high-quality reconstruction of the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of a flow chart of an image encryption and decryption method based on dynamic compressed sensing of the present invention;

[0023] Figure 2 It is a principle block diagram of an image encryption and decryption method based on dynamic compressed sensing of the present invention;

[0024] Figure 3 It is a schematic diagram of a reconstructed network model structure of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.

[0026] In view of the above problems, the present invention provides an image encryption and decryption method based on dynamic compressed sensing.

[0027] The present invention provides an image encryption and decryption method based on dynamic compressed sensing, comprising:

[0028] Step S1: input an original image and divide it into multiple sub-images, each sub-image has a size of N×N; in a certain embodiment, N=32.

[0029] Step S2: using a dynamic compression network to perform convolution compression and initialization reconstruction on the sub-image to obtain a preliminary encrypted image;

[0030] Step S3, using the SHA256 algorithm to calculate the 64-byte Hash value of the sub-image, split the Hash value with a step size of 2 bytes, use the two numbers of the first 4 bytes as the common factor α, and divide the last 60 bytes into β1 and β2; use α to perform XOR operations on β1 and β2 to obtain two 2-byte numbers, convert them into decimal and divide them by 255 to obtain two initial keys as the initial values ​​x0 and y0 of the two-dimensional Logistic chaotic system; due to the perturbation sensitivity of the SHA256 algorithm, ensure that one Figure 1 dense;

[0031] Step S4: The two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 to generate two chaotic sequences x n and n , from the chaotic sequence x n 32×2 numbers are cut out and enlarged and rounded to the interval [0,32]. The first 32 numbers are used for the row sliding scrambling of the preliminary encrypted image, and the last 32 numbers are used for the column sliding scrambling of the preliminary encrypted image. nTake 32×32×2 numbers, enlarge them to the interval [0, 255] and cut them into two parts as random keys for diffusion encryption; after sliding scrambling and diffusion encryption, the final encrypted ciphertext image is obtained;

[0032] Step S5, using two random sequences to decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain a decrypted image;

[0033] Step S6: reconstruct the decrypted image through a nonlinear deep reconstruction network to restore the sub-image;

[0034] Step S7: The sub-images are spliced ​​in the order of cutting to restore the original image.

[0035] Furthermore, the process of performing convolution compression and initialization reconstruction on the sub-image using the dynamic compression network specifically includes:

[0036] Step S21, using convolution kernel W com ∈R N×N×M (The convolutional layer does not set the activation function and bias) for the sub-image X∈R N ×N (N=32) Perform non-sliding convolution operation to achieve linear compression to obtain the observed signal

[0037] Step S22: Use convolution kernel (The convolutional layer does not set the activation function and bias) for the observed signal Convolution operation is performed to achieve linear initialization reconstruction. Where M is the number of channels, B 2 For quantity.

[0038] Furthermore, the step S3 comprises:

[0039] The sub-image is processed by SHA256 to obtain a 64-byte hash value, which is then divided into 32 2-byte values ​​in steps of 2 to obtain a sequence K. I ={k i |i∈[1,32],k i ∈[0,2 8 ]}.make As the public key factor, the remaining 30 k i Divide into 2 subsequences, XOR each subsequence with the public key factor to get two 2-byte numbers, and divide by 255 for normalization to get x0, y0; as shown in the following formula:

[0040]

[0041]

[0042] Furthermore, the two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 in the following way:

[0043] Take x0, y0 as initial values ​​and perform discrete iteration according to the following formula:

[0044] x n+1 =[4μ1x n (1-x n )+μ2x n y n ]mod 1

[0045] y n+1 =[4μ1y n (1-y n )+μ2x n y n ]mod 1

[0046] Among them, x n ,y n is the discrete iterative value of the two-dimensional Logistic chaotic map, μ1 and μ2 are the system parameters of the two-dimensional Logistic chaotic map, and x n y n is the secondary coupling term.

[0047] Furthermore, the sliding scrambling and diffusion process specifically includes:

[0048] Step C1: The preliminary encrypted image is first subjected to row sliding scrambling and then column sliding scrambling according to the following formula.

[0049] img rs [m i ,:]=roll(img[m i ,:],rs i )

[0050] img cs [:,n j ]=roll(img[:,n j ],cs j )

[0051] Among them, img rs [·]、img cs [·] represents the image after row and column sliding scrambling, roll(·) represents the sliding scrambling operation, img[·] represents the image before the scrambling operation (size is m×n), m i represents the i-th row of the image, n j represents the jth column of the image, rs i represents the sliding step length of the sliding scrambling of the i-th row, rs i∈[0,n-1],cs j represents the sliding step size of the jth column for sliding scrambling, cs j ∈[0,m-1].

[0052] Step C2: From the chaotic sequence y n Take 32×32×2 numbers, enlarge them to the interval [0, 255] and divide them into two parts as random keys K1 and K2 for diffusion encryption;

[0053] Step C3: The image obtained in step C1 is divided into rows and spliced ​​into a one-dimensional vector. First, it is subjected to backward XOR diffusion, that is, the i+1th pixel value is obtained by XORing the i-th pixel value and the i-th element of the key K1 (obtained in step S4). Further, the result vector of the backward XOR diffusion is subjected to forward XOR diffusion, that is, starting from the last pixel value, the i-th pixel value is obtained by XORing the i+1th pixel and the i+1th element of the key K2 (obtained in step S4).

[0054] Furthermore, the specific process of the nonlinear deep reconstruction network for reconstruction includes:

[0055] Step D1: Create a data set to perform network training on the reconstructed network model.

[0056] Step D2: Decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain a decrypted image as the input of the reconstruction network model (model training has been completed).

[0057] Step D3: The decrypted image undergoes convolution operations and nonlinear operations of the multi-layer convolutional layers of the reconstruction network model to obtain a reconstructed sub-image.

[0058] Embodiment 1:

[0059] An image encryption and decryption method based on dynamic compressed sensing, comprising:

[0060] Step 1: Input the original image and divide it into multiple sub-images, each sub-image has a size of N×N (N=32);

[0061] Step 2: Use the dynamic compression network to perform convolution compression and initialization reconstruction on the sub-image to obtain a preliminary encrypted image;

[0062] Step 3: Use the SHA256 algorithm to calculate the 64-byte Hash value of the sub-image, and divide the Hash value into 2-byte steps. The two numbers in the first 4 bytes are used as the common factor α, and the last 60 bytes are equally divided into β1 and β2. Use α to perform XOR operations on β1 and β2 to obtain two 2-byte numbers, convert them into decimal and divide them by 255 to obtain two initial keys as the initial values ​​x0 and y0 of the two-dimensional Logistic chaotic system. Since the SHA256 algorithm is sensitive to disturbances, it is guaranteed that Figure 1 dense;

[0063] Step 4: The two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 to generate two chaotic sequences. n 32×2 numbers are cut out and enlarged and rounded to the interval [0,32]. The first 32 numbers are used for the row sliding scrambling of the preliminary encrypted image, and the last 32 numbers are used for the column sliding scrambling of the preliminary encrypted image. n Take 32×32×2 numbers, enlarge them to the interval [0, 255] and cut them into two parts as random keys for diffusion encryption; after sliding scrambling and diffusion encryption, the final encrypted ciphertext image is obtained;

[0064] Step 5: Use two random sequences to decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain the decrypted image;

[0065] Step 6: The decrypted image is reconstructed through a nonlinear deep reconstruction network to restore the sub-image;

[0066] Step 7: The sub-images are spliced ​​in the order of cutting to restore the original image.

[0067] Furthermore, the process of performing convolution compression and initialization reconstruction on the sub-image using the dynamic compression network specifically includes:

[0068] Step 1: Use convolution kernel W com ∈R N×N×M (The convolutional layer does not set the activation function and bias) for the sub-image X∈R N×N (N=32) Perform non-sliding convolution operation to achieve linear compression to obtain the observed signal

[0069] Step 2: Use the convolution kernel (The convolutional layer does not set the activation function and bias) for the observed signal A convolution operation is performed to achieve linear initialization reconstruction.

[0070] Furthermore, the two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 in the following way:

[0071] Step 1: After the sub-image is processed by SHA256, a 64-byte hash value is obtained. The hash value is divided into 32 2-byte values ​​in steps of 2 to obtain a sequence K. I ={k i |i∈[1,32],k i ∈[0,2 8 ]}.make As the public key factor, the remaining 30 k i Divide into 2 subsequences, XOR each subsequence with the public key factor to get two 2-byte numbers, and divide by 255 for normalization to get x0, y0; as shown in the following formula:

[0072]

[0073]

[0074] Step 2: Use x0 and y0 as initial values ​​and perform discrete iterations according to the following formula:

[0075] x n+1 =[4μ1x n (1-x n )+μ2x n y n ]mod 1

[0076] y n+1 =[4μ1y n (1-y n )+μ2x n y n ]mod 1

[0077] Among them, x n ,y n is the discrete iterative value of the two-dimensional Logistic chaotic map, μ1 and μ2 are the system parameters of the two-dimensional Logistic chaotic map, and x n y n is the secondary coupling term.

[0078] Furthermore, the sliding scrambling and diffusion process specifically includes:

[0079] Step 1: Perform row sliding scrambling and column sliding scrambling on the preliminary encrypted image according to the following formula.

[0080] img rs [m i ,:]=roll(img[m i ,:],rs i )

[0081] img cs [:,nj ]=roll(img[:,n j ],cs j )

[0082] Among them, img rs [·]、img cs [·] represents the image after row and column sliding scrambling, roll(·) represents the sliding scrambling operation, img[·] represents the image before the scrambling operation (size is m×n), m i represents the i-th row of the image, n j represents the jth column of the image, rs i (rs i ∈[0,n-1]) represents the sliding step size of the sliding scrambling of the i-th row, cs j (cs j ∈[0,m-1]) represents the sliding step size for sliding scrambling of the j-th column.

[0083] Step 2: Divide the image obtained in step 1 into rows and concatenate them into a one-dimensional vector. First, perform backward XOR diffusion on it, that is, the i+1th pixel value is obtained by XORing the i-th pixel value and the i-th element of the key K1 (obtained in step 4). Further, the result vector of backward XOR diffusion is forward XOR diffused, that is, starting from the last pixel value, the i-th pixel value is obtained by XORing the i+1th pixel and the i+1th element of the key K2 (obtained in step 4).

[0084] Furthermore, the specific process of the nonlinear deep reconstruction network for reconstruction includes:

[0085] Step 1: Create a data set to perform network training on the reconstructed network model.

[0086] Step 2: Decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain the decrypted image as the input of the reconstruction network model (model training has been completed).

[0087] Step 3: The decrypted image undergoes convolution operations and nonlinear operations of the multi-layer convolutional layers of the reconstruction network model to obtain a reconstructed sub-image.

[0088] Embodiment 2:

[0089] The dataset uses 91-images 8-bit grayscale images as the reconstruction network training set, and the Set14 dataset 8-bit grayscale images as the test set. The pixel values ​​of the images in the dataset are compressed to the interval [0,1]. MSR-Net is used as the nonlinear deep reconstruction network. The system parameters of the two-dimensional Logistic chaotic system are set to μ1=0.88 and μ2=0.94. The training loss function of the network model uses the square loss function, the activation function is the ReLU function, the network layer weights are initialized with Gaussian random distribution, the adaptive learning rate adam is used, the initial value of the learning rate is 0.0001, and the number of iterations is 300. After the dynamic compression network and the reconstruction network are trained, according to Figure 1 and Figure 2 The forward encryption and decryption process of executing the network model is as follows:

[0090] Step 1: Take the test set image as the input original image and divide it into multiple sub-images, each of which is 32×32 in size;

[0091] Step 2: Use the dynamic compression network to perform convolution compression and initialization reconstruction on the sub-image to obtain a preliminary encrypted image;

[0092] Step 3: Use the SHA256 algorithm to calculate the 64-byte Hash value of the sub-image, divide the Hash value into 2-byte steps, use the two numbers in the first 4 bytes as the common factor α, and divide the remaining 60 bytes into β1 and β2; use α to perform XOR operations on β1 and β2 to obtain two 2-byte numbers, convert them into decimal and divide them by 255 to obtain two initial keys as the initial values ​​x0 and y0 of the two-dimensional Logistic chaotic system;

[0093] Step 4: The two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 to generate two chaotic sequences. n 32×2 numbers are cut out and enlarged to the interval [0,32]. The first 32 numbers are used for row sliding scrambling and the last 32 numbers are used for column sliding scrambling. n Take 32×32×2 numbers, enlarge them to the interval [0, 255] and divide them into two parts as random keys for diffusion encryption; after sliding scrambling and diffusion encryption, the final encrypted ciphertext image can be output;

[0094] Step 5: Use two random sequences to decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain the decrypted image;

[0095] Step 6: The decrypted image is reconstructed through a nonlinear deep network (such as Figure 3 As shown), reconstruct the sub-image;

[0096] Step 7: The sub-images are spliced ​​in the order of cutting to restore the original image.

[0097] Beneficial effects of the present invention:

[0098] (1) Reduced requirements for image sparsity during compression and encryption: The present invention uses a convolutional layer that removes activation functions and biases to achieve image compression, thereby ignoring whether the image is sparse;

[0099] (2) Utilizing the self-learning ability of deep convolutional neural networks to improve the problem of difficulty in achieving high-quality reconstruction at low sampling rates: The present invention obtains an image reconstruction network model that implements nonlinear operations by training a deep convolutional neural network. This model can obtain more feature information from low-sampling rate images, thereby achieving high-quality reconstruction of the image.

[0100] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An image encryption and decryption method based on dynamic compressed sensing, characterized in that: The method comprises the following steps: Step S1, input an original image and divide it into multiple sub-images, each sub-image has a size of N×N; Step S2: using a dynamic compression network to perform convolution compression and initialization reconstruction on the sub-image to obtain a preliminary encrypted image; Step S3, using the SHA256 algorithm to calculate the 64-byte Hash value of the sub-image, dividing the Hash value with a step length of 2 bytes, using the two numbers of the first 4 bytes as the common factor α, and dividing the remaining 60 bytes into β1 and β2; using α to perform an XOR operation on β1 and β2 to obtain two 2-byte numbers, converting them into decimal and dividing them by 255 to obtain two initial keys as the initial values ​​x0 and y0 of the two-dimensional Logistic chaotic system; Step S4: The two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 to generate two chaotic sequences x n and n , from the chaotic sequence x n 32×2 numbers are cut out and enlarged and rounded to the interval [0,32]. The first 32 numbers are used for row sliding scrambling of the preliminary encrypted image, and the last 32 numbers are used for column sliding scrambling of the preliminary encrypted image. Then from the chaotic sequence y n Take 32×32×2 numbers, enlarge them to the interval [0, 255] and divide them into two parts as the random key for diffusion encryption; After sliding scrambling and diffusion encryption, the final encrypted ciphertext image is obtained; Step S5, using two random sequences to decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain a decrypted image; Step S6, reconstructing the decrypted image through a nonlinear deep reconstruction network to restore the sub-image; Step S7: The sub-images are spliced ​​in the order of cutting to restore the original image.

2. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 1, characterized in that: N＝32。 3. The image encryption and decryption method based on dynamic compressed sensing according to claim 1 or 2, characterized in that: The method of using a dynamic compression network to perform convolution compression and initialization reconstruction on the sub-image includes: Step S21, using convolution kernel W com ∈R N×N×M For the sub-image X∈R N×N Perform non-sliding convolution operation to achieve linear compression to obtain the observed signal Step S22: Use convolution kernel For the observed signal Convolution operation is performed to achieve linear initialization reconstruction, where M is the number of channels and B is the number of channels. 2 For quantity.

4. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 3, characterized in that: The convolutional layer does not set the activation function and bias.

5. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 1, characterized in that: The step S3 comprises: The sub-image is processed by SHA256 to obtain a 64-byte hash value, which is then divided into 32 2-byte values ​​in steps of 2 to obtain a sequence K. I ={k i |i∈[1,32],k i ∈[0,2 8 ]};make As the public key factor, the remaining 30 k i Divide into 2 subsequences, XOR each subsequence with the public key factor to get two 2-byte numbers, and divide by 255 for normalization to get x0, y0; as shown in the following formula:

6. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 5, characterized in that: The two-dimensional Logistic chaotic system iterates the initial values ​​x0 and y0 to generate two chaotic sequences x n and n include: Take x0, y0 as initial values ​​and perform discrete iteration according to the following formula: x n+1 =[4μ1x n (1-x n )+μ2x n y n [mod 1 and n+1 =[4μ1y n (1-and n )+μ2x n and n ]mod 1 Among them, x n ,y n is the discrete iterative value of the two-dimensional Logistic chaotic map, μ1 and μ2 are the system parameters of the two-dimensional Logistic chaotic map, and x n y n is the secondary coupling term.

7. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 6, characterized in that: The system parameters of the two-dimensional Logistic chaotic system are set to μ1=0.88, μ2=0.

94.

8. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 6, characterized in that: The sliding scrambling and diffusion process in S4 specifically includes: Step C1: Perform row sliding scrambling and column sliding scrambling on the preliminary encrypted image according to the following formula; img rs [m i ,:]=roll(img[m i ,:],rs i ) img cs [:,n j ]=roll(img[:,n j ],cs j ) Among them, img rs [·], img cs [·] represents the image after row and column sliding scrambling, roll(·) represents the sliding scrambling operation, img[·] represents the image before the scrambling operation, m i represents the i-th row of the image, n j represents the jth column of the image, rs i represents the sliding step length of the sliding scrambling of the i-th row, rs i ∈[0,n-1],cs j represents the sliding step size of the jth column for sliding scrambling, cs j ∈[0,m-1]; Step C2: From the chaotic sequence y n Take 32×32×2 numbers, enlarge them to the interval [0, 255] and divide them into two parts as random keys K1 and K2 for diffusion encryption; Step C3: Divide the image obtained in step C1 by rows and concatenate them into a one-dimensional vector; first perform backward XOR diffusion on it, that is, the i+1th pixel value is obtained by XORing the i-th pixel value and the i-th element of the key K1; further, the result vector of the backward XOR diffusion is forward XOR diffused, that is, starting from the last pixel value, the i-th pixel value is obtained by XORing the i+1th pixel and the i+1th element of the key K2.

9. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 8, characterized in that: The reconstruction by the nonlinear deep reconstruction network comprises: Step D1: Create a data set to perform network training on the reconstructed network model; Step D2: Decrypt the ciphertext image using inverse XOR diffusion and inverse sliding scrambling to obtain a decrypted image as the input of the reconstruction network model; Step D3: The decrypted image undergoes convolution operations and nonlinear operations of the multi-layer convolutional layers of the reconstruction network model to obtain a reconstructed sub-image.

10. The image encryption and decryption method based on dynamic compressed sensing as claimed in claim 9, characterized in that: The nonlinear deep reconstruction network uses MSR-Net.

Images