A method for enhancing the consistency of carrier image content for large-scale steganography

Abstract

The present invention proposes a method for enhancing the content consistency of carrier images for large-scale steganography. The method involves splitting a color carrier image into multiple channels and generating an embedding probability map using a generative network. A noise map is then used to embed messages, and the embedding probability is iteratively updated to minimize the difference between the steganographic image and the carrier image. A noise mixing weight matrix and a two-class flow noise strategy are further used to generate a noisy image. A diffusion model combined with a regularizer guides the denoising process, ultimately generating a large-scale enhanced carrier image with consistent content. The enhanced image generated by the present invention can significantly increase its embedding capacity while maintaining content consistency with the original carrier image, thereby producing a steganographic image with even better steganographic performance.

Description

Technical Field

[0001] The present invention relates to the technical field of image steganography, and in particular to a method for enhancing the consistency of carrier image content for large-capacity steganography. Background Art

[0002] With the widespread penetration of the internet, the security of private image information faces severe challenges and has become a critical issue that demands urgent attention. Against this backdrop, leveraging precisely adapted steganography technology to protect private image information has become a top priority in the information security field. As a powerful weapon in information security, steganography's core advantage lies in its ability to cleverly conceal the traces of "covert communications" and establish secure communication links in a manner that is difficult to detect.

[0003] Currently, most steganography algorithms can indeed embed secret information into a selected carrier image without the user noticing. However, these algorithms often find it difficult to achieve an ideal balance between embedding capacity and security. The texture complexity of carrier images varies greatly, which means that different images have different embedding potentials. Specifically, when the carrier image is an ordinary image, no matter how sophisticated the objective distortion function is designed, existing methods find it difficult to find a satisfactory balance between embedding capacity and security. Images with simple textures can only accommodate limited secret information. If the embedding capacity is forcibly increased, it is very easy to cause visual distortion, thereby exposing the steganographic traces and reducing security. Although images with complex textures appear to have a larger embedding space, they also face the problem of how to accurately grasp the embedding depth during the embedding process to avoid destroying the image structure due to excessive embedding, which can then be detected by the detection algorithm. Summary of the Invention

[0004] In response to the above problems, the present invention proposes a method for enhancing the content consistency of carrier images for large-capacity steganography, which can enhance the embedding capacity of the original carrier image and generate a large-capacity enhanced carrier image with consistent content. It is used to solve the technical problem that the carrier image selected by the existing image steganography method is usually far from the optimal option for embedding secret information.

[0005] The specific scheme of the present invention is as follows:

[0006] A method for enhancing the consistency of carrier image content for large-capacity steganography includes the following steps:

[0007] S1, split the color carrier image dataset into R, G, B multi-channel format and feed it to the generation network G N Get the multi-channel embedding probability map M T p , and then embed the probability graph M T p With noise map N TSend it to the embedding simulator E for message embedding to obtain the modified mapping M T ;

[0008] S2: By modifying the mapping M T Obtain a stego-image, use a steganalyzer to minimize the difference between the stego-image and the carrier image, and iteratively update the embedding probability of each pixel in the multi-channel;

[0009] S3: Embedding probability map M using multiple channels T p Calculate the noise mixing weight matrix W used to enhance the image N ;

[0010] S4: A dual-class noise strategy is adopted for the carrier image, and the noise mixing weight matrix W N To guide the adaptive mixing, the noisy image C is obtained t0 ;

[0011] S5: C t0 It is fed into the diffusion model and constrained by three regularizers: image sharpness, noise distribution, and adversarial deformation to guide the denoising process and generate the final denoised image C0.

[0012] Furthermore, the color carrier image dataset is split into R, G, B multi-channel formats and fed to the generation network G in S1. N Get the multi-channel embedding probability map M T p , specifically including the following steps:

[0013] S11, take out the color carrier images in the data set in batches of batch_size, and perform channel splitting on the carrier images in each batch to obtain The number of grayscale images C T ;

[0014] S12, will The number of grayscale images C T Feed to the generator network G N and ensure that the generator network can output the corresponding The number of multi-channel embedding probability maps M T p , T∈{R,G,B}.

[0015] By splitting the color carrier image into channels and processing them in batches, we can efficiently generate multi-channel embedding probability maps. This processing method not only improves the accuracy of the generated embedding probability maps, but also ensures stable output of the generated network under a fixed payload, providing a reliable foundation for subsequent message embedding and image optimization.

[0016] Furthermore, in S1, the probability map M is embedded T p With noise map N T Send it to the embedding simulator E for message embedding to obtain the modified mapping M T , specifically including the following steps:

[0017] S13, embed the probability map M T p Each color carrier image corresponds to 3 different channel embedding probability maps and 3 sets of random noise mapping N T Send them together to the embedding simulator E for message embedding;

[0018] S14, the embedding simulator E simulates the embedding process by sampling using pixel-level modification, mapping random noise to N T and the embedding probability map M T p By comparison, we obtain a pixel-level modification map M with three possible directions (+1, -1, 0) T .

[0019] By combining the embedding probability map with the noise map and simulating the embedding process through pixel-level modification sampling, we can obtain a precise pixel-level modification map. This precise modification map helps better control image changes when embedding information, reduces the impact on image quality, and improves the stealth and security of information embedding.

[0020] Furthermore, the embedding rules of the embedding simulator E are as follows:

[0021] M T p (x,y) direction (+1) and M T p The embedding probability of direction (-1) in (x,y) is M T p (x,y) / 2;

[0022] If a random noise mapping element N T (x,y) is less than M T p When the embedding probability of the direction (+1) in (x,y) is increased, the mapping M is modified. T (x,y)=+1;

[0023] If a random noise mapping element N T (x,y) is greater than M T p (x,y) (embedding probability of 1-direction (-1)), then modify the mapping M T (x,y)=-1;

[0024] Otherwise, set M T (x,y)=0.

[0025] The embedding simulator uses embedding rules based on probability comparison, enabling more flexible simulation of the information embedding process and ensuring that the embedding operation conforms to a predetermined probability distribution. This rule clarifies the decision-making basis for pixel modification directions, helping to improve the accuracy and controllability of information embedding while reducing unnecessary image modifications, thereby better maintaining image quality.

[0026] Furthermore, in S2, by modifying the mapping M T Obtain the stego image, including:

[0027] By modifying the mapping M T Compared with the previously separated R, G, B multi-channel grayscale image C T The steganographic image S can be obtained T :

[0028] .

[0029] By combining a modified mapping with a multi-channel grayscale image to generate a stego image, the embedded information can be directly reflected in the image. This method of generating stego images is simple and efficient, ensuring the accuracy of the embedded information and the recoverability of the image, providing a good foundation for subsequent steganalysis and image optimization.

[0030] Furthermore, in S2, a steganalyzer is used to minimize the difference between the stego-image and the carrier image, specifically comprising the following steps:

[0031] S21, the steganalyzer detects the embedded information in the input steganalyzer image and obtains the embedding probability;

[0032] S22, selecting a sample image in the next iteration from the stego image according to the embedding probability.

[0033] By detecting embedded information and selecting sample images, the steganalyzer effectively evaluates the quality of the embedded image and the effectiveness of information embedding. This mechanism selects appropriate images for the next round of iterative optimization based on the embedding probability, gradually improving the similarity between the stego-image and the original carrier image, enhancing the concealment and security of steganography.

[0034] Furthermore, in S3, the multi-channel embedding probability map M is used T p Calculate the noise mixing weight matrix W used to enhance the image N Specifically include:

[0035] S31, uses the embedding probability generator model to generate a multi-channel embedding probability map by inputting a color image;

[0036] S32, binarize the multi-channel embedding probability map to obtain a new binary embedding probability image B T ;

[0037] S33, embedding probability image B according to the binary T , for any position (x,y), if B T There is at least one value of 1 in (x, y), then in the weight matrix W N In the , set the value of the position to 1, otherwise set it to 0.

[0038] By using an embedded probability generator model and binarization to calculate the noise mixing weight matrix, we can accurately determine the areas in the image suitable for noise mixing. This binarization-based method simplifies the weight matrix generation process while effectively highlighting the noise mixing weights of important areas in the image, providing precise guidance for subsequent adaptive mixing.

[0039] Furthermore, in S4, a dual-class flow noise strategy is adopted for the carrier image, specifically including:

[0040] The two-type flow noise mechanism includes creative flow and stable flow;

[0041] For the creative flow, random noise corresponding to time step t is added to the carrier image C, thus generating a noisy image C O t ;

[0042] Using the diffusion model to analyze C O t Perform iterative denoising until the time step t0 is reached, at which point the denoised image C is obtained. O t0 ;

[0043] For stable flow, DDIM inversion is used to add noise to the carrier image C and obtain the noisy image C S t0 .

[0044] A dual-stream noise strategy is employed, generating noise images with different characteristics through a creative stream and a stable stream. This mechanism enhances image detail and creativity while maintaining image content fidelity. The creative stream adds random noise to enrich image detail and improve image embedding capacity, while the stable stream adds noise through DDIM inversion to maintain content fidelity, providing a diverse foundation for subsequent adaptive blending and optimization of images.

[0045] Furthermore, in S4, the noise mixing weight matrix W N To guide the adaptive mixing, the noisy image C is obtained t0 , specifically including the following steps:

[0046] When the dual-class flow noise generates two noise images C O t0 and C S t0 Then, C is dynamically adjusted by embedding the probability value in the probability image O t0 and C S t0 The usage ratio is calculated using the weight matrix W N Adaptively blend two noisy images;

[0047] For image regions with higher embedding probability, C S t0 Maintain a relatively stable content structure. For image areas with low embedding probability, use C O t0 Generate variants to enrich the detail information in this area.

[0048] Adaptive blending, guided by a noise blending weight matrix, dynamically adjusts the proportion of noise images used. This adaptive blending method flexibly selects noise images based on the characteristics of the image region, ensuring stability in areas with high embedding probability while enriching detailed information in areas with low embedding probability, thereby increasing the embedding capacity in these areas. This sophisticated blending process helps significantly improve the overall embedding capacity of the carrier image while maintaining the consistency of the carrier image content.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] A universal content-consistent carrier image enhancement framework is proposed, which can enhance the embedding capacity of any given image or image set while maintaining content consistency. By automatically learning the embedding cost of each pixel in the image, an embedding probability map is generated to evaluate the embedding ability of each pixel. Then, the mixing ratio of the two-class flow noise is dynamically adjusted according to the embedding probability map to obtain a noisy image. The noisy image is then fed into a diffusion model constrained by three regularizers for denoising, and the denoised enhanced image is output. Finally, the large-capacity enhanced image with improved embedding capacity is used as a carrier image, which can be embedded with secret messages using any feasible steganography method for secure communication in the network. It has been proven that the proposed framework is effective for the most advanced steganography methods in enhancing the embedding capacity of the carrier image, so the proposed framework has good versatility.

[0051] A multi-channel embedding probability joint learning mechanism is proposed. This mechanism optimizes performance by comprehensively learning all channel information of color images. During the learning process, the embedding probability information between different channels is shared and coordinated, ensuring in-depth learning and iterative updating of the embedding probability information of different channels for the same detailed parts of the image, generating an embedding probability map with deeper spatial information and finer coarse-grainedness. This mechanism effectively solves the problems of difficult-to-control inter-channel coordination and color continuity faced when directly learning embedding probabilities for color images, thereby avoiding color distortion or the generation of small colored dots. At the same time, it also overcomes the color disharmony and image quality degradation caused by the inability to capture image color information when converting color images to grayscale images for embedding probability learning.

[0052] (3) A two-class stream noise dynamic mixing strategy based on multi-channel embedding probability map is designed. This strategy can use the multi-channel embedding probability map to calculate the noise mixing weight matrix for image enhancement, and then use the weight matrix as a guide to adaptively mix noise to generate noisy images. In this process, for image areas with higher embedding probability, we use stable noise to maintain a stable content structure. For image areas with lower embedding probability, we use creative noise generation variants to enrich the detail information in the area. This strategy significantly enhances the overall embedding capacity of the image by enhancing the texture details of areas that are not suitable for embedding. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present drawings or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present drawings. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0054] Figure 1 is a flow chart of the method of the present invention;

[0055] Figure 2 A schematic diagram of visualization of the original carrier image and the enhanced carrier image provided by an embodiment of the present invention;

[0056] Figure 3 A schematic diagram comparing the embedding probabilities of the original carrier image and the enhanced carrier image provided by an embodiment of the present invention;

[0057] Figure 4 A schematic diagram comparing the embedding probabilities of the original carrier image and the enhanced carrier image provided by an embodiment of the present invention;

[0058] Figure 5A schematic diagram comparing the embedding probabilities of the original carrier image and the enhanced carrier image provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention. Based on the embodiments provided by the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0060] like Figure 1 As shown, the present invention provides a method for enhancing the consistency of carrier image content for large-capacity steganography, which specifically includes the following steps:

[0061] S1, split the color carrier image dataset into R, G, B multi-channel format and feed it to the generation network G N Get the multi-channel embedding probability map M T p , and then embed the probability graph M T p With noise map N T Send it to the embedding simulator E for message embedding to obtain the modified mapping M T .

[0062] Specifically, the color carrier image dataset is split into R, G, B multi-channel formats and fed into the generation network G N Get the multi-channel embedding probability map M T p , specifically including the following steps:

[0063] S11, take out the color carrier images in the data set in batches of batch_size, and perform channel splitting on the carrier images in each batch to obtain The number of grayscale images C T ;

[0064] S12, will The number of grayscale images C T Feed to the generator network G N and ensure that the generator network can output the corresponding The number of multi-channel embedding probability maps M T p , T∈{R,G,B}.

[0065] In the specific implementation process, since this framework is a general framework, any feasible generator network can refer to the generator network G N, such as HILL, UT-GAN, ASDL-GAN and SPAR-RL, can be used for this operation.

[0066] Specifically, embed the probability map M T p With noise map N T Send it to the embedding simulator E for message embedding to obtain the modified mapping M T , specifically including the following steps:

[0067] S13, embed the probability map M T p Each color carrier image corresponds to 3 different channel embedding probability maps and 3 sets of random noise mapping N T Send them together to the embedding simulator E for message embedding;

[0068] In the embedding probability map M T p In the example, each color carrier image corresponds to three different channel embedding probability maps, namely M R p 、M G p and M B p , and map it with 3 sets of random noise N T Send them together to the embedding simulator E for message embedding, and get the modified mapping M T :

[0069] .

[0070] S14, the embedding simulator E simulates the embedding process by sampling using pixel-level modification, mapping random noise to N T and the embedding probability map M T p By comparison, we obtain a pixel-level modification map M with three possible directions (+1, -1, 0) T .

[0071] In the specific implementation process, SPAR-RL can be selected as the generation network G N .

[0072] In addition, the embedding rule logic of the embedding simulator E is as follows:

[0073] M T p (x,y) direction (+1) and M T p The embedding probability of direction (-1) in (x,y) is M T p (x,y) / 2;

[0074] If a random noise mapping element N T (x,y) is less than M T p When the embedding probability of the direction (+1) in (x,y) is increased, the mapping M is modified. T (x,y)=+1;

[0075] If a random noise mapping element N T (x,y) is greater than M T p When (1-(embedding probability of direction (-1)) in (x,y), the mapping M is modified T (x,y)=-1;

[0076] Otherwise, set M T (x,y)=0.

[0077] S2, by modifying the mapping M T A steganalyzer is used to minimize the difference between the steganalyzer and the carrier image, and the embedding probability of each pixel in the multi-channel is iteratively updated to obtain a more accurate multi-channel embedding probability map.

[0078] Specifically, by modifying the mapping M T Obtain the stego image, including:

[0079] By modifying the mapping M T Compared with the previously separated R, G, B multi-channel grayscale image C T The steganographic image S can be obtained T :

[0080] .

[0081] During the specific implementation process, modify the mapping M T The element type in is (+1, -1, 0), which can be compared with the corresponding grayscale image C according to the channel type. T Perform matrix element addition processing. Taking a single channel as an example, if there is a grayscale image matrix [[119, 20, 98], [248, 6, 58]] and a modified mapping matrix [[-1, -1, 0], [1, -1, 0]], then the processed stego image matrix is ​​[[118, 19, 98], [249, 5, 58]].

[0082] Specifically, a steganalyzer is used to minimize the difference between the stego-image and the carrier image, which specifically includes the following steps:

[0083] S21, the steganalyzer detects the input steganalyzer image S T The embedded information in is used to obtain the embedding probability.

[0084] The main task of the steganalyzer is to detect the input steganalyzer image S T Whether it contains embedded information, such as Xu-Net, Yedroudj-Net, Ye-Net and SRNet, can be used for this operation.

[0085] S22, according to the embedding probability from the stego image S T Select the image to sample in the next iteration.

[0086] The steganalysis image S that performs better during steganalysis T There will be a greater probability of being sampled in the next round of iteration. Specifically, if the value of the embedding probability is higher, the corresponding pixel is suitable for hiding the secret message, that is, its embedding ability is higher, and vice versa. The purpose is to minimize the stego-image S T and the original image C T The gap between them is narrowed and the embedding probability of each pixel in the multi-channel is iteratively updated to obtain a more accurate multi-channel embedding probability map.

[0087] S3, using multi-channel embedding probability map M T p Calculate the noise mixing weight matrix W used to enhance the image N .

[0088] Specifically, the following steps are included:

[0089] S31, uses the embedding probability generator model to generate a multi-channel embedding probability map by inputting a color image.

[0090] S32, binarize the multi-channel embedding probability map to obtain a new binary embedding probability image B T .

[0091] Specifically, set a threshold , use the threshold decision function to obtain the binary embedding probability image B T :

[0092] ;

[0093] Among them, (x, y) is the coordinate position of a probability value embedded in the probability map.

[0094] In the specific implementation process, taking a single channel as an example, if there is an embedding probability map [[0.11, 0.48, 0.36], [0.25, 0.25, 0.43]], and the threshold is taken When it is 0.4, the new binary embedding probability image after processing by the threshold decision function is [[0, 1, 0], [0, 0, 1]].

[0095] S33, embedding probability image B according to the binary T , for any position (x,y), if B T There is at least one value of 1 in (x, y), then in the weight matrix W N In the example, the value of the position is set to 1, otherwise it is set to 0. The process is described by the following formula:

[0096] .

[0097] In the specific implementation process, if there is a binary embedding probability image B T is {[[0, 0, 1], [0, 0, 0]], [[1, 0, 1], [1, 0, 0]], [[0, 0, 1], [1, 0, 1]]}, then according to the above weight matrix W N The calculation rules show that the weight matrix W N is [[1, 0, 1], [1, 0, 1]].

[0098] S4 adopts a dual-class noise strategy for the carrier image and uses the noise mixing weight matrix W N To guide the adaptive mixing, the noisy image C is obtained t0 .

[0099] Specifically, a dual-class flow noise strategy is adopted for the carrier image, including:

[0100] The dual-stream noise mechanism includes creative stream and stable stream. The creative stream aims to enhance the richness and creativity of image details by incorporating higher-intensity noise elements, while the stable stream focuses on introducing relatively weak noise to ensure that the fidelity of the content is not significantly affected.

[0101] For the creative flow, random noise corresponding to time step t is added to the carrier image C, thus generating a noisy image C O t , the process can be described by the following formula:

[0102] ;

[0103] Among them, α t Represents the weight coefficient, which is used to regulate the contribution of the carrier image C;

[0104] σ t Indicates the noise intensity;

[0105] and represents the random noise vector at time step t.

[0106] Using the diffusion model to analyze C O tPerform iterative denoising until the time step t0 is reached, at which point the denoised image C is obtained. O t0 To ensure that the structural elements of the image are consistent with the carrier image C, that is, to maintain content consistency, gradient guided sampling is used to introduce the conditional reflection of auxiliary information for C in the noise flow. O t →C O t0 denoising process.

[0107] In the specific implementation process, the basic model of Stable Diffusion XL (SDXL-base) implemented in the HuggingFace Transformer and Diffuser library is used as the diffusion model for image enhancement.

[0108] For stable flow, DDIM inversion is used to add noise to the carrier image C and obtain the noisy image C S t0 It ensures that when using a deterministic sampling algorithm like DDIM, S t0 Reconstruct the content of the carrier image C with high fidelity.

[0109] Specifically, the noise mixing weight matrix W N To guide the adaptive mixing, the noisy image C is obtained t0 , specifically including the following steps:

[0110] When the dual-class flow noise generates two noise images C O t0 and C S t0 Then, C is dynamically adjusted by embedding the probability value in the probability image O t0 and C S t0 The usage ratio is calculated using the weight matrix W N Adaptively blend two noisy images;

[0111] Its function expression is:

[0112] .

[0113] For image regions with higher embedding probability, C S t0 Maintain a relatively stable content structure. For image areas with low embedding probability, use C O t0 Generate variants to enrich the detail information in this area.

[0114] In the specific implementation process, if there is a weight matrix W N For [[1, 0, 1], [1, 0, 1]], the noise image C O t0 [[0.878, 0.011, 0.596], [0.336, 0.121, 0.633]], noisy image C S t0 [[0.256, 0.894, 0.290], [0.632, 0.158, 0.534]], then the stable noise C is dynamically adjusted according to the above adaptive mixed noise function S t0 and create noise C O t0 After using the ratio, we can get the noisy image C t0 is [[0.878, 0.894, 0.596], [0.336, 0.158, 0.633]].

[0115] S5, C t0 It is fed into the diffusion model and constrained by three regularizers: image sharpness, noise distribution, and adversarial deformation to guide the denoising process and generate the final denoised image C0.

[0116] Specifically, in the C t0 In the process of image denoising, three target attributes are formulated as constraints from the score-based perspective of the diffusion model, and they are used to guide the sampling process by regulating the predicted noise, thereby adjusting the output. t0 Assigned to C t .

[0117] Sharpness regularization,Sharpness refers to the perceived clarity related to the edge contrast of an image.,Due to the characteristics of the human visual system, images with higher,sharpness tend to appear clearer, although an increase in sharpness does not necessarily,improve the actual resolution of the image.

[0118] Here, use The sharpening rate is used to regularize denoising, Represents the intermediate reconstruction of C0 at time step t using the reparameterization technique, i.e. Represents the intermediate reconstruction version of image C0 when time step t is close to 0. Specifically, the Sobel kernel is used to estimate the The magnitude of the spatially varying brightness derivative is expressed as ; In order to encourage higher sharpness and improve the overall generation quality, a binary indicator ∨ ​​(·) is used for optimization. When the input value falls within The 35th and 65th percentiles When ∨(·)=1, the goal of sharpness regularization is:

[0119] ;

[0120] in, represents the sharpness regularization loss;

[0121] W represents the height and width of the image;

[0122] The sum is normalized to make it independent of the image size, and a negative sign is added to indicate that minimizing this loss encourages higher sharpness;

[0123] It means to sum all pixel positions (x, y) of the image.

[0124] Distribution regularization, taking into account the necessity of generalization error, diffusion model The predicted noise may not follow a Gaussian distribution N(0, I), especially when the diffusion model is directly used to generate images from synthetic noise images produced in the noise stage. Therefore, the denoising process is regularized by penalizing the distribution gap:

[0125] ;

[0126] in, It is used to calculate the variance of the prediction noise;

[0127] Represents the L2 norm, which is used to measure the difference between two vectors;

[0128] represents the distribution regularization loss.

[0129] Adversarial Regularization, driven by the self-attention guidance of the diffusion model, adversarial regularization is added in the denoising stage to avoid blurry images. Specifically, Defined as a Gaussian blur function, and the target is designed as:

[0130] ;

[0131] in, represents the adversarial regularization loss;

[0132] Reconstruct the intermediate image Apply a Gaussian blur function ,This step will generate a blurred version of the image;

[0133] Compute the difference between the original intermediate reconstructed image and the blurred image.

[0134] With the help of these three regularizations, an additional operation step is set after each denoising iteration to update the current state:

[0135] ;

[0136] in, The image state after regularization optimization at time step t-1;

[0137] Represents the image state at time step t-1;

[0138] C t Represents the image state at time step t;

[0139] ▽ is the gradient operator used to represent the gradient information, (ξ, ψ, ζ) is the trade-off parameter used to control the relative importance of the regularization loss;

[0140] When the time step t iterates to 0, the final denoised image C0 can be generated.

[0141] In the specific implementation process, the optimal value of the trade-off parameter is determined through experimental research, (ξ, ψ, ζ) = (4, 20, 0.4), and the visualization diagram of the original carrier image C and the final enhanced carrier image C0 is compared with the embedding probability diagram as shown in the figure below. Figure 2-Figure 5 shown.

[0142] The following is an analysis of various aspects of the performance of the embodiments of the present invention:

[0143] Peak signal-to-noise ratio (PSNR) analysis: PSNR is one of the commonly used indicators for measuring image quality. It is used to compare the quality difference between the original image and the image after encoding, compression or processing. The higher the PSNR value, the better the image quality. Its values ​​can be roughly divided into four categories: (1) [0,20) dB, unacceptable image quality; (2) [20,30) dB, poor image quality, noticeable; (3) [30,40) dB; acceptable quality but distortion may be noticeable; (4) [40, ) dB, the image quality is excellent (that is, very close to the original image), and the PSNR value V can be quantitatively calculated using the following formula PSNR :

[0144] ;

[0145] ;

[0146] Among them G (u,v) , Z (u,v)Represent the pixel values ​​at coordinates (u, v) of the original image and the stego-image respectively;

[0147] V MES is the mean square error, which is used to quantify the degree of pixel-level distortion between the original image and the stego-image.

[0148] Structural Similarity Index (SSIM) analysis: SSIM is an indicator used to measure the similarity between two images. Its value range is between -1 and 1, where 1 means that the two images are exactly the same and -1 means that the two images are completely different. The SSIM value can be quantitatively calculated using the following formula: :

[0149] ;

[0150] Where, µ p and µ q Represent the average brightness of the original image G and the stego image Z respectively;

[0151] σ p and σ q represent the standard deviations of G and Z, respectively;

[0152] σ pq represents the covariance of G and Z;

[0153] C1 and C2 are constants to avoid instability when the denominator is close to 0.

[0154] Experimental Dataset: The Human Preference Dataset, Version 2 (HPDv2) is a large-scale dataset of human preferences covering images generated by a wide range of text prompts. It includes 433,760 human preference choices on 798,090 pairs of images and provides a set of evaluation prompts, which are evenly distributed according to four major style categories: animation, concept art, painting, and photography. For each evaluation prompt type, HPDv2 provides corresponding benchmark images generated by various mainstream text-to-image generation models. On this basis, this paper adopts a set of benchmark images generated by the SDXL-Base-0.9 model as the original image set for image enhancement technology, which contains 10,000 color images.

[0155] Steganography scheme: SPAR-RL is selected as the steganography scheme of this embodiment, and steganographic images of the original image and the enhanced image with a payload of 0.1bpp to 0.4bpp are generated. In addition, the multi-channel joint learning mechanism proposed in the present invention is used for color image processing.

[0156] The analysis and calculation results of the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) of different stego images are shown in Table 1 below:

[0157]

[0158] exist Figure 2 In the figure, the original carrier image and the enhanced carrier image are visually compared, and the detailed texture is magnified. The first row shows the unprocessed original images, and the second row shows the carefully designed enhanced images. From the perspective of the overall content information of the image, the enhanced carrier image strictly maintains the consistency of content with the original carrier image, without introducing any changes that deviate from the original content theme. The subject objects and information presented in the two images are completely consistent, ensuring the faithful reproduction of the content. Secondly, in the detailed texture part, by observing the area marked by the box in the figure, it can be clearly seen that in the enhanced image, this area not only reveals more delicate and complex texture information, but also perfectly integrates the generated information that is highly consistent with the content, thereby significantly improving the image's detail performance and information communication capabilities. Therefore, it is verified that this framework can perfectly generate richer, more delicate and highly integrated texture information in the details while ensuring the consistency of the image content before and after enhancement.

[0159] In order to further observe the effect of improving the embedding ability of carrier images, Figure 3-Figure 5 The embedding probability maps of the original carrier image and the enhanced carrier image are shown in the figure. The first row is the original image and the enhanced image, and the second row is the corresponding embedding probability map. The higher the pixel brightness in the map, the stronger the embedding ability. By comparison, it can be clearly observed that after enhancing the original carrier image, the pixel brightness in the larger box area of ​​the embedding probability map changes from darker to brighter, indicating that the embedding ability of this area has been enhanced. Since the embedding ability of the original carrier image is immutable under the same load, the message can only be embedded in pixels with a lower embedding probability. When the embedding ability of the larger box area of ​​the enhanced image is improved, the choice of message embedding is transferred from the smaller box area to this area, resulting in the pixel brightness in the smaller box area changing from bright to dim. This phenomenon shows that the enhanced image can carry additional secret messages with a higher embedding probability. It is verified that the large-capacity enhanced image generated by the present invention can effectively improve the embedding ability of the original carrier image at the same embedding cost.

[0160] It should be noted that the present invention is not limited to the above-mentioned embodiments. The above-mentioned embodiments are merely examples, and any embodiments having substantially the same structure and effect as the technical concept within the scope of the technical solution of the present invention are all included in the technical scope of the present invention. In addition, without departing from the scope of the present invention, other embodiments that can be conceived by those skilled in the art and that combine some of the constituent elements in the embodiments are also included in the scope of the present invention.

Claims

1. A method for enhancing the consistency of carrier image content for large-capacity steganography, characterized in that: The following steps are involved: S1, split the color carrier image dataset into R, G, B multi-channel format and feed it to the generation network G N Get the multi-channel embedding probability map M T p , and then embed the probability graph M T p With noise map N T Send it to the embedding simulator E for message embedding to obtain the modified mapping M T ; S2: By modifying the mapping M T Obtain a stego-image, use a steganalyzer to minimize the difference between the stego-image and the carrier image, and iteratively update the embedding probability of each pixel in the multi-channel; S3: Embedding probability map M using multiple channels T p Calculate the noise mixing weight matrix W used to enhance the image N ; S4: A dual-class noise strategy is adopted for the carrier image, and the noise mixing weight matrix W N To guide the adaptive mixing, the noisy image C is obtained t0 The dual-class stream noise strategy is adopted for the carrier image, specifically including: The two-type flow noise mechanism includes creative flow and stable flow; For the creative flow, random noise corresponding to time step t is added to the carrier image C, thus generating a noisy image C O t ; Using the diffusion model to analyze C O t Perform iterative denoising until the time step t0 is reached, at which point the denoised image C is obtained. O t0 ; For stable flow, DDIM inversion is used to add noise to the carrier image C and obtain the noisy image C S t0 ; S5: C t0 It is fed into the diffusion model and constrained by three regularizers: image sharpness, noise distribution, and adversarial deformation to guide the denoising process and generate the final denoised image C0.

2. A method for enhancing the consistency of carrier image content for large-capacity steganography according to claim 1, characterized in that: In S1, the color carrier image dataset is split into R, G, B multi-channel formats and fed into the generation network G N Get the multi-channel embedding probability map M T p , specifically including the following steps: S11, take out the color carrier images in the data set in batches of batch_size, and perform channel splitting on the carrier images in each batch to obtain The number of grayscale images C T ; S12, will The number of grayscale images C T Feed to the generator network G N and ensure that the generator network can output the corresponding The number of multi-channel embedding probability maps M T p , T∈{R,G,B}.

3. The method for enhancing the consistency of carrier image content for large-capacity steganography according to claim 1, characterized in that: In S1, the probability map M is embedded T p With noise map N T Send it to the embedding simulator E for message embedding to obtain the modified mapping M T , specifically including the following steps: S13, embed the probability map M T p Each color carrier image corresponds to 3 different channel embedding probability maps and 3 sets of random noise mapping N T Send them together to the embedding simulator E for message embedding; S14, the embedding simulator E simulates the embedding process by sampling using pixel-level modification, mapping random noise to N T and the embedding probability map M T p By comparison, we obtain a pixel-level modification map M with three possible directions (+1, -1, 0) T .

4. A method for enhancing the consistency of carrier image content for large-capacity steganography according to claim 3, characterized in that: The embedding rules of the embedding simulator E are as follows: M T p (x,y) direction (+1) and M T p The embedding probability of direction (-1) in (x,y) is M T p (x,y) / 2; If a random noise mapping element N T (x,y) is less than M T p When the embedding probability of the direction (+1) in (x,y) is increased, the mapping M is modified. T (x,y)=+1; If a random noise mapping element N T (x,y) is greater than M T p When the embedding probability of 1-direction (-1) in (x,y) is changed, the mapping M is modified. T (x,y)=-1; Otherwise, set M T (x,y)=0.

5. The method for enhancing the consistency of carrier image content for large-capacity steganography according to claim 2, characterized in that: In S2, by modifying the mapping M T Obtain the stego image, including: By modifying the mapping M T Compared with the previously separated R, G, B multi-channel grayscale image C T The steganographic image S can be obtained T : 。 6. The method for enhancing the consistency of carrier image content for large-capacity steganography according to claim 1, characterized in that: In S2, a steganalyzer is used to minimize the difference between the stego-image and the carrier image, specifically comprising the following steps: S21, the steganalyzer detects the embedded information in the input steganalyzer image and obtains the embedding probability; S22, selects a sample image in the next iteration from the stego image according to the embedding probability.

7. The method for enhancing the consistency of carrier image content for large-capacity steganography according to claim 1, characterized in that: In S3, the multi-channel embedding probability map M is used T p Calculate the noise mixing weight matrix W used to enhance the image N Specifically include: S31, uses the embedding probability generator model to generate a multi-channel embedding probability map by inputting a color image; S32, binarize the multi-channel embedding probability map to obtain a new binary embedding probability image B T ; S33, embedding probability image B according to the binary T , for any position (x,y), if B T There is at least one value of 1 in (x, y), then in the weight matrix W N In the , set the value of the position to 1, otherwise set it to 0.

8. The method for enhancing the consistency of carrier image content for large-capacity steganography according to claim 7, characterized in that: In S4, the noise mixing weight matrix W N To guide the adaptive mixing, the noisy image C is obtained t0 , specifically including the following steps: When the dual-class flow noise generates two noise images C O t0 and C S t0 Then, C is dynamically adjusted by embedding the probability value in the probability image O t0 and C S t0 The usage ratio is calculated using the weight matrix W N Adaptively blend two noisy images; For image regions with higher embedding probability, C S t0 Maintain a relatively stable content structure. For image areas with low embedding probability, use C O t0 Generate variants to enrich the detail information in this area.

Images