Image steganography method and system based on dynamic latent variable and frequency domain embedding

Abstract

The invention relates to the technical field of image encryption, and discloses an image steganography method and system based on dynamic latent variables and frequency domain embedding, and the method comprises the steps: obtaining a hash value and a plurality of secret keys of a to-be-encrypted color plaintext image; constructing a hyperchaotic system, and iteratively generating a plurality of key streams; a to-be-encrypted color plaintext image is decomposed into R, G and B channel matrixes, and three ciphertext sequences are obtained based on the three channel matrixes; the method comprises the following steps: initializing a noise image, determining a time step by using a secret key, and executing reverse denoising of a denoising diffusion implicit model (DDIM) with a fixed step length to obtain an intermediate latent variable; based on the intermediate latent variable and the three ciphertext sequences, obtaining a dense frequency domain-containing sub-band, and performing inverse discrete wavelet transform on the low-frequency sub-band and the dense frequency domain-containing sub-band to obtain a dense latent variable; and executing DDIM denoising of the time step by taking the secret-containing latent variable as a starting point to obtain a secret-containing image. By adopting the method, the security and visual quality of image steganography can be improved, and high embedding capacity is ensured.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of image encryption, in particular to an image steganography method and system based on dynamic latent variables and frequency domain embedding. BACKGROUND

[0002] Information transmission in an open network is threatened by theft and tampering. Traditional cryptography can encrypt content, but the obvious "ciphertext" it produces can expose the communication behavior itself, which is easy to intercept. Image steganography technology aims to embed secret information in public images unnoticeably, making the carrier image look natural statistically, and achieving covert communication. However, early methods (such as LSB replacement, modifying DCT / DWT coefficients) use fixed rules to embed information. This changes the statistical characteristics of the image. With the development of deep learning-based steganalysis technology, these statistical anomalies become extremely easy to detect, and the security is low. To improve concealment, researchers turn to GAN, normalized flow (NF) and other generative models to directly synthesize images containing secrets. However, this method has two major problems: it is difficult to guarantee both high visual quality of the image containing secrets and high capacity of the embedded information, and the information is usually fixedly injected into a specific location of the model. This lack of variation limits security.

[0003] In the prior art, diffusion models provide a new idea for steganography due to their high-quality image generation capability and quasi-reversible generation process. Existing research attempts to use them for steganography. However, existing methods mainly use them as a whole generator, the information injection point is single, and the key problem is that they fail to fully utilize the dynamic change of the internal state of the diffusion model over time steps in the step-by-step generation process. This neglect of the internal dynamic characteristics results in an embedding strategy that is not flexible and concealed enough, hindering further improvement in security. SUMMARY

[0004] The purpose of the present application is to solve at least one of the defects existing in the prior art, and to provide an image steganography method and system based on dynamic latent variables and frequency domain embedding, which can improve the security and visual quality of image steganography and ensure high embedding capacity.

[0005] To achieve the above-mentioned purpose, in a first aspect, the application provides an image steganography method based on dynamic latent variables and frequency domain embedding, which comprises the following steps: obtaining the hash value of a color plaintext image to be encrypted, and obtaining a plurality of keys according to the hash value; constructing a hyperchaotic system, initializing the hyperchaotic system using the plurality of keys, and iteratively generating a plurality of key streams; decomposing the color plaintext image to be encrypted into R, G, B three channel matrices, performing scrambling and diffusion operations on each channel respectively, and then performing value range expansion to obtain three ciphertext sequences; Initialize a noise image satisfying a standard Gaussian distribution, and determine a time step using the key, perform fixed-step DDIM inverse denoising to obtain an intermediate latent variable, the fixed step being the difference between a preset time and the time step; Discrete wavelet transform is performed on three channels of the intermediate latent variable respectively to obtain frequency domain subbands, and the three ciphertext sequences are respectively embedded into specified decimal places of non-low frequency subbands of the corresponding color channels to obtain ciphertext-containing frequency domain subbands, the frequency domain subbands including low frequency subbands and three high frequency subbands; Inverse discrete wavelet transform is performed on the low frequency subbands and the ciphertext-containing frequency domain subbands to obtain a ciphertext-containing latent variable; DDIM denoising of the time step is performed with the ciphertext-containing latent variable as the starting point to obtain a ciphertext-containing image.

[0006] Further, the hash value of the color plaintext image to be encrypted is obtained, and a plurality of keys are obtained according to the hash value, specifically including: The hash value of the color plaintext image to be encrypted is calculated using a hash function, the hash value is divided into a plurality of groups according to a first predetermined number of digits, and each group of hash values after grouping is converted into a decimal number; The obtained decimal number is divided into a plurality of groups according to a second predetermined number of digits, and the decimal numbers after grouping are summed to obtain the sum result of each group; The sum result of each group is multiplied by after adding 1 to obtain a plurality of keys, the range of the keys is , and the specific calculation formula is as follows:

[0007] wherein, .

[0008] Further, the mathematical model of the hyperchaotic system is as follows:

[0009] wherein, , , , , are parameters of the chaotic system.

[0010] Further, the hyperchaotic system is initialized using the plurality of keys to iteratively generate a plurality of key streams, specifically including: substituting the plurality of keys as initial values into the hyperchaotic system, iteratively calculating to obtain new initial values until an iteration threshold is reached to stop, and outputting a plurality of key streams.

[0011] Further, the color plaintext image to be encrypted is decomposed into R, G, BThree channel matrices, after performing scrambling and diffusion operations on each channel respectively, value range expansion is performed to obtain three ciphertext sequences, specifically including: 1). The color plaintext image to be encrypted is decomposed into R, G, B Three channel matrices, each channel matrix is flattened to obtain three one-dimensional sequences; 2). Three scrambling sequences are generated using the plurality of key streams, and the three one-dimensional sequences are scrambled using the three scrambling sequences to obtain three scrambled one-dimensional sequences; 3). Cross-channel mixed diffusion is performed on the three scrambled one-dimensional sequences to obtain initial ciphertext sequences, wherein the cross-channel mixed diffusion includes initialization and iterative diffusion; 4). The initial ciphertext sequence is mapped to obtain a value range expanded ciphertext sequence; 5). And each value in the value range expanded ciphertext sequence is decomposed into three-dimensional decimal representation and then concatenated in turn to obtain the final three ciphertext sequences.

[0012] Further, when scrambling the three one-dimensional sequences using the three scrambling sequences, scrambling is performed at least 3 times.

[0013] Further, a noise image satisfying a standard Gaussian distribution is initialized, and DDIM reverse denoising with a fixed step size is performed using the key to determine the time step to obtain an intermediate latent variable, specifically including: Initialize a random noise image subject to a standard Gaussian distribution ; Set the interval , convert it to the parameter as follows:

[0014] Wherein is a key related to the plaintext, denotes a modulo operation; Take as the initial value, perform several steps of DDIM reverse denoising to obtain an intermediate latent variable , and the iterative formula is as follows:

[0015] Wherein, , , , , , , , is the predicted noise obtained by the diffusion model.

[0016] Further, the three channels of the intermediate latent variable are respectively subjected to discrete wavelet transform to obtain frequency domain subbands, and the three ciphertext sequences are respectively embedded into specified decimal places of non-low frequency subbands of corresponding color channels to obtain ciphertext-containing frequency domain subbands, specifically including: The intermediate latent variable is subjected to three color channel separation, and each channel is subjected to discrete wavelet transform to obtain frequency domain subbands, the frequency domain subbands including low frequency subbands and three high frequency subbands; For each color channel, the low frequency subbands are kept unchanged, and the three ciphertext sequences are sequentially embedded into specified decimal places of the three high frequency subbands in a preset order to obtain three ciphertext-containing frequency domain subbands.

[0017] Further, the embedding calculation method is as follows:

[0018] Among them, is a floating-point wavelet coefficient in the selected high frequency subband, is a single secret information number to be embedded obtained from the ciphertext sequence , is the number of bits of the replaced decimal place, is a sign function.

[0019] In a second aspect, the present application also provides an image steganography system based on dynamic latent variable and frequency domain embedding, the system being based on the method of the first aspect and including the following modules: An acquisition module configured to acquire a hash value of a color plaintext image to be encrypted, and obtain a plurality of keys according to the hash value; A construction module configured to construct a hyperchaotic system, initialize the hyperchaotic system using the plurality of keys, and iteratively generate a plurality of key streams; A scrambling and diffusion module configured to decompose the color plaintext image to be encrypted into R, G, B three channel matrices, perform scrambling and diffusion operations on each channel respectively, and then perform value range expansion to obtain three ciphertext sequences; An initialization module configured to initialize a noise image satisfying a standard Gaussian distribution, determine a time step using the keys, and perform fixed-step DDIM reverse denoising to obtain an intermediate latent variable, the fixed step being a difference between a preset time and the time step; A discrete wavelet transform module configured to respectively perform discrete wavelet transform on three channels of the intermediate latent variable to obtain frequency domain subbands, and respectively embed the three ciphertext sequences into specified decimal places of non-low frequency subbands of corresponding color channels to obtain ciphertext-containing frequency domain subbands, the frequency domain subbands including low frequency subbands and three high frequency subbands; inverse discrete wavelet transform module: configured to perform inverse discrete wavelet transform on the low-frequency subband and the secret-containing frequency domain subband to obtain a secret-containing latent variable; de-noising module: starting from the secret-containing latent variable, performing DDIM de-noising of the time step to obtain a secret-containing image.

[0020] Compared with the prior art, the present application has the following beneficial effects: The present application randomizes the embedding positions of different images by dynamically associating the plaintext hash value with the embedding time step, solves the defect that the traditional static embedding is easy to be detected, effectively improves the image steganography security; the secret information is converted into a noise-like sequence by the chaotic system, while ensuring security, effectively reduces the regularity disturbance to the generation process, thereby improving the visual quality of image steganography; a high-frequency subband decimal digit replacement strategy is also used, which ensures the visual quality while realizing high-capacity embedding. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flowchart of a dynamic latent variable and frequency domain embedding-based image steganography method according to Embodiment 1 of the present application; Figure 2 A block diagram of a dynamic latent variable and frequency domain embedding-based image steganography system according to Embodiment 3 of the present application; Figure 3 A color plaintext image to be encrypted according to Embodiment 2 of the present application, wherein, Figure 3 a is a color plaintext House image to be encrypted; Figure 3 b is a color plaintext Baboon image to be encrypted; Figure 3 c is a color plaintext Plane image to be encrypted; Figure 3 d is a color plaintext Peppers image to be encrypted; Figure 4 A recovered image according to Embodiment 2 of the present application, wherein, Figure 4 a is a recovered image of a ciphertext House image; Figure 4 b is a recovered image of a ciphertext Baboon image; Figure 4 c is a recovered image of a ciphertext Plane image; Figure 4 d is a recovered image of a ciphertext Peppers image; Figure 5 An extraction fidelity diagram of a decrypted image under different embedding time steps; Figure 6Illustration of generated image quality and imperceptibility at different embedding time steps. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and not to limit the present application.

[0023] Embodiment 1 Please refer to Figure 1 The image steganography method based on dynamic latent variable and frequency domain embedding according to the embodiment of the present application preferably comprises the following steps: S1: obtaining a hash value of a color plaintext image to be encrypted, and obtaining a plurality of keys according to the hash value; S2: constructing a hyper-chaotic system, initializing the hyper-chaotic system using the plurality of keys, and iteratively generating a plurality of key streams; S3: decomposing the color plaintext image to be encrypted into R, G, B three channel matrices, performing scrambling and diffusion operations on each channel respectively, and then performing value range expansion to obtain three ciphertext sequences; S4: initializing a noise image satisfying a standard Gaussian distribution, determining a time step using the keys, performing fixed-step DDIM (denoising diffusion implicit model) reverse denoising to obtain an intermediate latent variable, and the fixed step being a difference between a preset time and the time step; S5: performing discrete wavelet transform on three channels of the intermediate latent variable to obtain frequency domain subbands, and embedding the three ciphertext sequences into specified decimal places of non-low frequency subbands of the corresponding color channels to obtain ciphertext-containing frequency domain subbands, the frequency domain subbands including low frequency subbands and three high frequency subbands; S6: performing inverse discrete wavelet transform on the low frequency subbands and the ciphertext-containing frequency domain subbands to obtain a ciphertext-containing latent variable; S7: taking the ciphertext-containing latent variable as a starting point, performing DDIM denoising for the time step to obtain a ciphertext-containing image.

[0024] In this embodiment, the hash value of the plaintext and the embedding time step are dynamically associated, so that the embedding positions of different images are completely randomized, solving the defect that the traditional static embedding is easy to be detected, and effectively improving the image steganography security; the chaotic system is used to convert secret information into a noise-like sequence, which effectively reduces the regularity disturbance on the generation process while ensuring security, thereby improving the visual quality of image steganography; and the high frequency subband decimal replacement strategy is also used, which ensures the visual quality while realizing high capacity embedding.

[0025] In an optional embodiment, in step S1, specifically comprising: A hash value of the color plaintext image to be encrypted is calculated using a hash function, the hash value is divided into multiple groups according to a first predetermined number of bits, and each group of the divided hash value is converted into a decimal number; it should be noted that the hash function includes SHA-256, SHA-3, etc., and the present embodiment does not limit which function is used to process the color plaintext image; The obtained decimal number is divided into multiple groups according to a second predetermined number of bits, and the grouped decimal numbers are summed to obtain a summation result of each group; The summation result of each group is multiplied by after being added by 1, to obtain multiple keys, the range of the keys is within , and the specific calculation formula is as follows:

[0026] wherein, .

[0027] For example, a color plaintext image to be encrypted with a size of is obtained, and a SHA-256 hash function is used to calculate the corresponding 256-bit hash value of the color plaintext image ; the 256-bit hash value is grouped with 8 bits as a group, a total of 32 groups, each group is converted into a decimal number, and is denoted as to ; the to are evenly divided into 4 groups, each group contains 8 integers, the sum of the integers in each group is calculated, and then the sum value is divided by 255 and taken modulo 1, thereby calculating four keys with a range within .

[0028] In an optional embodiment, in step S2, the mathematical model of the hyperchaotic system is as follows:

[0029] wherein, , , , , are parameters of the chaotic system, in the present embodiment, , , , , The multiple keys are used to initialize the hyperchaotic system, and multiple key streams are iteratively generated, specifically including: the multiple keys are substituted as initial values into the hyperchaotic system, new initial values are iteratively calculated until the iteration threshold is reached to stop, and multiple key streams are output.

[0030] For example, four keys are substituted into the hyperchaotic system as initial values, iterated and discarded the first 1000 iterations to eliminate transient effects, generating four random sequences of length , , , , , .

[0031] In an optional embodiment, in step S3, specifically includes: S3.1: the color plaintext image to be encrypted is decomposed into R, G, B three channel matrices, each channel matrix is flattened to obtain three one-dimensional sequences, this step converts two-dimensional image data into one-dimensional sequences to facilitate subsequent scrambling and diffusion operations; S3.2: three scrambling sequences are generated using the multiple key streams, and the three one-dimensional sequences are scrambled using the three scrambling sequences to obtain three scrambled one-dimensional sequences, scrambled at least three times, this step generates random scrambling sequences through chaotic key streams to prepare for subsequent pixel scrambling, and scrambles at least three times to enhance the confusion effect and disrupt the position of pixels within each channel, destroy the spatial correlation of the image, and enhance security; S3.3: cross-channel mixed diffusion is performed on the three scrambled one-dimensional sequences to obtain an initial ciphertext sequence, wherein the cross-channel mixed diffusion includes initialization and iterative diffusion, this step realizes the diffusion between pixels by performing XOR operation on the current pixel value, the encryption result of the previous time and the key stream. At the same time, the encryption results of the three channels affect each other (cross use the previous state), enhancing the complexity and security of encryption; S3.3: mapping is performed on the initial ciphertext sequence to obtain a ciphertext sequence with expanded value range; S3.4: and each value in the ciphertext sequence with expanded value range is decomposed into three decimal representations and then concatenated in turn to obtain three final ciphertext sequences, this step expands the encrypted 8-bit integer (0-255) to a wider value range (0-999), and then decomposes it into three digits (0-9), which not only increases the capacity of subsequent embedding (because each pixel is expanded to three digits), but also makes it more natural when embedding into the decimal place of wavelet coefficients (because each digit is 0-9, which can be embedded into a specific position after the decimal point).

[0032] For example, 1). a color plaintext image with a size of is decomposed into Decomposed into , and three color channel matrix, the size of the channel matrix is also , each matrix is flattened into a one-dimensional sequence, denoted as , , ; 2). Combined key and generate a permutation sequence from to , similarly, combined key and generate permutation sequence , combined key stream and generate permutation sequence ; in this embodiment, the pseudo code for generating permutation sequence , , is as follows:

[0033] 3). Permute the pixel sequence of each channel, and repeat the permutation operation three times to enhance the confusion effect. The permuted pixel sequence is denoted as , , , the specific calculation formula is as follows

[0034] wherein , .

[0035] 4). Mix and diffuse the three channels. First, initialize the first element of each channel using the permuted pixel sequence. The specific initialization operation is as follows:

[0036] 5). Perform iterative diffusion operation on the subsequent elements of each channel. The calculation method of iterative diffusion is as follows:

[0037] wherein , , , is the encrypted ciphertext sequence.

[0038] 6). Through mapping, the original value domain of each element of the ciphertext sequence is expanded to ​Each extended integer is decomposed into its three-digit decimal representation (with leading zeros if the number is less than three digits), and concatenated in order to form a ciphertext sequence of three digits from 0 to 9 , , , both of which have a length of .

[0039] In an optional embodiment, in step S4, specifically comprises: S4.1: initialize a random noise image subject to a standard Gaussian distribution , with a size of ; in this step, the size of the noise image is set to , which is four times the size of the plaintext image I, which can provide sufficient space capacity for frequency domain embedding.

[0040] S4.2: set the interval , and convert it to the parameter as follows:

[0041] wherein is a key related to the plaintext, denotes a modulo operation; S4.3: perform several steps of DDIM inverse denoising with as the initial value to obtain an intermediate latent variable . The iteration formula is as follows:

[0042] wherein , , , , , , , is the predicted noise obtained by the diffusion model.

[0043] In an optional embodiment, in step S5, specifically comprises: S5.1: perform three-color channel separation on the intermediate latent variable, and perform discrete wavelet transform on each channel to obtain frequency domain subbands, the frequency domain subbands including a low-frequency subband and three high-frequency subbands; S5.2: for each color channel, keep the low-frequency subband unchanged, and embed the three ciphertext sequences in the specified decimal places of the three high-frequency subbands in a preset order to obtain three frequency domain subbands containing the ciphertext. The calculation method of embedding is as follows:

[0044] wherein, is a floating-point wavelet coefficient in a selected high-frequency subband, is a single secret information number to be embedded from the ciphertext sequence , is the number of decimal digits to be replaced, is a sign function.

[0045] Illustratively, 1). The intermediate latent variable is subjected to three-color channel separation, and a two-dimensional discrete wavelet transform is applied to each color channel to obtain four frequency-domain subbands , , and of equal size .

[0046] 2). For each color channel, the low-frequency subband is kept completely unchanged, and a ciphertext sequence is sequentially embedded in the coefficients of the three high-frequency subbands , , in a predetermined order to obtain three high-frequency subbands , , that have carried secret information. The specific embedding operation is decimal digit replacement, as follows:

[0047] wherein, is a floating-point wavelet coefficient in a selected high-frequency subband, is a single secret information number to be embedded from the ciphertext sequence , is the number of decimal digits to be replaced, is a sign function.

[0048] 3). The high-frequency subbands , , that contain secret information in each color channel are subjected to a two-dimensional inverse discrete wavelet transform with the unmodified low-frequency subband , and the three channels are combined to obtain a latent variable that contains secret information.

[0049] In an optional embodiment, in step S6, specifically comprising: the high-frequency subbands , , that contain secret information in each color channel are subjected to a two-dimensional inverse discrete wavelet transform with the unmodified low-frequency subband , and the three channels are combined to obtain a latent variable that contains secret information.

[0050] In an optional embodiment, in step S7, specifically, starting from , the remaining DDIM denoising process is continued until , and the final output image is the stego image carrying the secret information.

[0051] In this embodiment, by dynamically associating the plaintext hash value with the embedding time step, the embedding positions of different images are completely randomized, solving the defect that traditional static embedding is easy to be detected, effectively improving the security of image steganography; by converting the secret information into a noise-like sequence through a chaotic system, the security is ensured while effectively reducing the regularity disturbance to the generation process, thereby improving the visual quality of image steganography; and by using a high-frequency sub-band decimal digit replacement strategy, high-capacity embedding is achieved while ensuring the visual quality.

[0052] Embodiment 2 This embodiment is a simulation experiment of the method proposed in Embodiment 1, specifically, Figure 3 is the color plaintext image to be encrypted, wherein Figure 3 a is the color plaintext House image to be encrypted; Figure 3 b is the color plaintext Baboon image to be encrypted; Figure 3 c is the color plaintext Plane image to be encrypted; Figure 3 d is the color plaintext Peppers image to be encrypted; after encryption by the method of Embodiment 1, decryption is performed to obtain the recovered Figure 4 , wherein Figure 4 a is the recovered image of the ciphertext House image; Figure 4 b is the recovered image of the ciphertext Baboon image; Figure 4 c is the recovered image of the ciphertext Plane image; Figure 4 d is the recovered image of the ciphertext Peppers image; In addition, assuming that the total diffusion step length is , this embodiment first generates a key using the hash value of the plaintext image, and calculates a specific embedding time step within a relatively optimal window . The performance trade-off of this process is shown in Figure 5 , 6 Subsequently, starting from a random Gaussian noise , the intermediate latent variable is obtained by denoising to time . At this time, the processed secret information is embedded in the latent variable to obtain the latent variable containing the secret and the remaining denoising steps are continued to finally generate a high-quality stego image. During decryption, the same key is used to re-calculate the time steps , and the stego image is recovered to state by the DDIM forward noise adding process , and the secret information can be extracted therefrom to complete decryption.

[0053] Embodiment 3 Referring to Figure 2 , the image steganography system based on dynamic latent variables and frequency domain embedding according to an embodiment of the present application comprises the following modules: An acquisition module configured to acquire a hash value of a color plaintext image to be encrypted, and obtain a plurality of keys according to the hash value; A construction module configured to construct a hyperchaotic system, initialize the hyperchaotic system using the plurality of keys, and iteratively generate a plurality of key streams; A permutation and diffusion module configured to decompose the color plaintext image to be encrypted into three channel matrices R, G, and B, perform permutation and diffusion operations on each channel respectively, and then perform value range expansion to obtain three ciphertext sequences; An initialization module configured to initialize a noise image satisfying a standard Gaussian distribution, determine a time step using the keys, perform fixed-step DDIM reverse denoising to obtain an intermediate latent variable, and the fixed step is the difference between a preset time and the time step; A discrete wavelet transform module configured to perform discrete wavelet transform on three channels of the intermediate latent variable respectively to obtain frequency domain subbands, and embed the three ciphertext sequences into specified decimal places of non-low frequency subbands of the corresponding color channels to obtain ciphertext-containing frequency domain subbands, the frequency domain subbands including low frequency subbands and three high frequency subbands; An inverse discrete wavelet transform module configured to perform inverse discrete wavelet transform on the low frequency subbands and the ciphertext-containing frequency domain subbands to obtain a ciphertext-containing latent variable; A denoising module configured to perform DDIM denoising of the time step using the ciphertext-containing latent variable as a starting point to obtain a ciphertext-containing image.

[0054] The system proposed in this embodiment is based on the method described in Embodiment 1, therefore, the optional items proposed in Embodiment 1 are applicable to this embodiment, to avoid repetition, therefore, they will not be described here.

[0055] Obviously, the above embodiments of the present application are merely exemplary but not intended to limit the embodiments of the present application. Based on the above description, any other variations or changes can be made by those skilled in the art without departing from the spirit and principles of the present application. It is not necessary to list all the embodiments here. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall fall within the scope of the claims of the present application.

Claims

1. An image steganography method based on dynamic latent variables and frequency domain embedding, characterized in that, The method comprises the following steps: Obtain the hash value of the color plaintext image to be encrypted, and obtain a plurality of keys according to the hash value; Construct a hyperchaotic system, initialize the hyperchaotic system using the keys, and iteratively generate a plurality of key streams; The color plaintext image to be encrypted is decomposed into R, G, B three channel matrices, and scrambling and diffusion operations are performed on each channel respectively, and then value range expansion is performed to obtain three ciphertext sequences; Initialize a noise image satisfying a standard Gaussian distribution, determine a time step using the keys, perform fixed-step DDIM reverse denoising, and obtain an intermediate latent variable, wherein the fixed step is the difference between a preset time and the time step; Discrete wavelet transform is performed on the three channels of the intermediate latent variable respectively to obtain frequency domain subbands, and the three ciphertext sequences are respectively embedded in the specified decimal places of the non-low frequency subbands of the corresponding color channels to obtain ciphertext-containing frequency domain subbands, wherein the frequency domain subbands include low frequency subbands and three high frequency subbands; Inverse discrete wavelet transform is performed on the low frequency subbands and the ciphertext-containing frequency domain subbands to obtain a ciphertext-containing latent variable. The ciphertext-containing latent variable is taken as a starting point, and DDIM denoising of the time step is performed to obtain a ciphertext-containing image.

2. The method of claim 1, wherein, The obtaining of the hash value of the color plaintext image to be encrypted and the obtaining of a plurality of keys according to the hash value specifically comprise: A hash function is used to calculate the hash value of the color plaintext image to be encrypted, the hash value is divided into a plurality of groups according to a first predetermined number of digits, and each group of hash values after grouping is converted into a decimal number; The obtained decimal numbers are divided into a plurality of groups according to a second predetermined number of digits, and the decimal numbers after grouping are summed to obtain the sum of each group. The sum of each group is multiplied by , and the result is added to 1 to obtain a plurality of keys, the range of which is , and the specific calculation formula is as follows: wherein .

3. The method of claim 1, wherein, The mathematical model of the hyperchaotic system is as follows: wherein, , , , , are parameters of the chaotic system.

4. The method of claim 3, wherein, The hyperchaotic system is initialized using the plurality of keys, and a plurality of key streams are iteratively generated, specifically comprising: substituting the plurality of keys as initial values into the hyperchaotic system, iteratively calculating new initial values until the iteration threshold is reached to stop, and outputting a plurality of key streams.

5. The method according to any one of claims 1 to 4, characterized in that, The color plaintext image to be encrypted is decomposed into R, G, B three channel matrices, and scrambling and diffusion operations are performed on each channel respectively, and then value range expansion is performed to obtain three ciphertext sequences. 1). decompose the color plaintext image to be encrypted into R, G, B three channel matrices, flatten each channel matrix to obtain three one-dimensional sequences; 2). Three scrambling sequences are generated using the plurality of key streams, and the three one-dimensional sequences are scrambled using the three scrambling sequences to obtain three scrambled one-dimensional sequences; 3). Cross-channel mixed diffusion is performed on the three scrambled one-dimensional sequences to obtain initial ciphertext sequences, wherein the cross-channel mixed diffusion includes initialization and iterative diffusion; 4). The initial ciphertext sequences are mapped to obtain ciphertext sequences with expanded value ranges; 5). Each value in the ciphertext sequences with expanded value ranges is decomposed into a three-dimensional decimal representation and then concatenated in sequence to obtain the final three ciphertext sequences.

6. The method of claim 5, wherein, When the three one-dimensional sequences are scrambled using the three scrambling sequences, at least 3 scramblings are performed.

7. The method of claim 1, wherein, A noise image satisfying a standard Gaussian distribution is initialized, a time step is determined using the keys, and fixed-step DDIM reverse denoising is performed to obtain an intermediate latent variable, specifically comprising: initializing a random noise image subject to a standard Gaussian distribution ; Setting interval Converting this to parameters As follows: wherein is a key related to the plaintext; With as initial values, perform several steps of DDIM inverse denoising to obtain an intermediate latent variable with the following iterative formula: wherein , , , , , , , is the predicted noise from the diffusion model.

8. The method of claim 1, wherein, Discrete wavelet transform is performed on the three channels of the intermediate latent variable respectively to obtain frequency domain subbands, and the three ciphertext sequences are respectively embedded in the specified decimal places of the non-low frequency subbands of the corresponding color channels to obtain ciphertext-containing frequency domain subbands, specifically comprising: The intermediate latent variable is subjected to three-color channel separation, and each channel is subjected to discrete wavelet transform to obtain a frequency domain subband, the frequency domain subband including a low-frequency subband and three high-frequency subbands; For each color channel, the low-frequency subband is kept unchanged, and the three ciphertext sequences are sequentially embedded into designated decimal places of the three high-frequency subbands in a preset order to obtain three frequency domain subbands containing the ciphertext.

9. The method of claim 8, wherein, The embedding calculation is as follows: wherein, is a floating-point wavelet coefficient in a selected high-frequency subband, is a single secret information number to be embedded from a ciphertext sequence , is the number of bits of a decimal digit to be replaced, is a sign function.

10. An image steganography system based on dynamic latent variables and frequency domain embedding, characterized in that, The system based on the method of any one of claims 1-9 includes the following modules: An acquisition module configured to acquire a hash value of a color plaintext image to be encrypted, and obtain a plurality of keys according to the hash value; A construction module configured to construct a hyperchaotic system, initialize the hyperchaotic system using the plurality of keys, and iteratively generate a plurality of key streams; scrambling and diffusion module: configured to decompose the color plaintext image to be encrypted into R, G, and B three channel matrices, respectively perform scrambling and diffusion operations on each channel, and then perform range expansion to obtain three ciphertext sequences; An initialization module configured to initialize a noise image satisfying a standard Gaussian distribution, determine a time step using the key, perform fixed-step DDIM reverse denoising, and obtain an intermediate latent variable, the fixed step being a difference between a preset time and the time step; A discrete wavelet transform module configured to respectively perform discrete wavelet transform on three channels of the intermediate latent variable to obtain a frequency domain subband, and embed the three ciphertext sequences into designated decimal places of non-low-frequency subbands of the corresponding color channels to obtain a frequency domain subband containing the ciphertext, the frequency domain subband including a low-frequency subband and three high-frequency subbands; An inverse discrete wavelet transform module configured to perform inverse discrete wavelet transform on the low-frequency subband and the frequency domain subband containing the ciphertext to obtain a ciphertext-containing latent variable; A denoising module configured to perform DDIM denoising of the time step using the ciphertext-containing latent variable as a starting point to obtain a ciphertext-containing image.

Images