A color image steganography method based on a diffusion model
By using a color image steganography method based on a diffusion model, the problems of small steganography capacity, poor robustness, and high computational overhead in existing technologies are solved, achieving high-security and high-quality secret information transmission, which is suitable for secret information transmission over public channels.
Patent Information
- Application Number
- CN202410028434.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-09
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-01-09
AI Technical Summary
Existing steganography algorithms have limited steganography capacity and poor robustness on grayscale images. Steganography of natural images is difficult to obtain pixel distribution. Convolutional neural network steganography has high computational overhead and limited image quality and size, making it difficult to meet the requirements of high security and high payload.
A color image steganography method based on a diffusion model is adopted. Through preprocessing, generating image pixels and probability distributions, transforming and embedding secret information, a high-quality large-size image is generated using a diffusion model, and the secret information is securely transmitted through STC encoding.
It improves the message carrying capacity and security of encrypted images, generates high-quality encrypted images with strong resistance to steganalysis, and is suitable for transmitting secret information through public channels.
Smart Images

Figure CN117834786B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of information security, specifically relating to a color image steganography method based on a diffusion model to enable the transmission of confidential messages over public channels. Background Technology
[0002] Steganography, as an important means of covert communication, not only fulfills the function of encrypting messages in traditional cryptography but also achieves the security of the behavior of both communicating parties. Current steganography algorithms have evolved from the initial LSB algorithm to subsequent algorithms such as HUGO, WOW, and HILL proposed within the minimum distortion framework. Although these algorithms have considered issues such as the embedding location of secret information and payload distribution, natural steganography, based on the 'cover-source' transformation idea of natural images, is more attractive in terms of security. Furthermore, with the development of neural networks, convolutional neural networks have also been used for steganography tasks, such as HiDDeN and HiNet. Diffusion models, as an emerging and powerful generative model, have great potential in terms of generation quality, sample diversity, and generation speed. Compared to adversarial generative networks, the rigorous mathematical theory makes it possible to obtain the sample probability distribution during the generation process, making its application to steganography tasks possible. Although using neural networks has brought significant performance improvements to steganography algorithms, there are still problems that need to be solved.
[0003] Current traditional steganography algorithms and neural network steganography algorithms have the following drawbacks:
[0004] 1. Traditional steganography algorithms based on the minimum distortion framework and STC encoding are mainly designed for grayscale images, and have drawbacks such as limited steganography capacity and poor robustness.
[0005] 2. Natural image steganography based on the source transformation concept has high security, but it is generally difficult to obtain the pixel distribution of the image, so its application is not widespread and the payload is small.
[0006] 3. Steganography algorithms using convolutional neural networks require the simultaneous design of encoders and decoders for joint training, which makes the network difficult to converge and results in high computational overhead.
[0007] 4. Steganography algorithms based on generative models suffer from poor image quality, small image size, and low carrying capacity of secret information in the generated images. Summary of the Invention
[0008] The present invention addresses the shortcomings of the existing technology by proposing a color image steganography method based on a diffusion model. This method aims to increase the message carrying capacity of the encrypted image while adding more secret messages, and to ensure the security of the encrypted image, thereby achieving higher image quality.
[0009] The present invention adopts the following technical solution to solve the technical problem:
[0010] The color image steganography method based on a diffusion model of the present invention is characterized by the following steps:
[0011] Step 1: Preprocessing of confidential information;
[0012] Step 1.1: After encrypting the confidential information using an encryption mechanism, an encrypted sequence is obtained;
[0013] Step 1.2: Use a converter to convert the encrypted sequence into a binary sequence m;
[0014] Step 2: Obtain the pixel count and probability distribution of the generated image;
[0015] Step 2.1, from the normal distribution Randomly sample a×b×c values and construct the input image x. T Where a and b are the height and width of the input image, and c is the number of channels of the input image;
[0016] Step 2.2: Input image x T The input is processed in the diffusion model, and the generated image x0 is output.
[0017] Using the G1 algorithm on x T The process yields the pixels and their distribution in the generated image x0. The set of pixels in the generated image x0 is denoted as X = {x}. i,j |i=1,2,…,a;j=1,2,…,b}, where x i,j Let X represent the pixel in the i-th row and j-th column; let Y = {Y} be the set of pixel distributions in X. i,j |i=1,2,…,a;j=1,2,…,b};Y i,j Represents the pixel x in the i-th row and j-th column. i,j The distribution, and In this context, ~ indicates obedience. x represents i,j The mean, x represents i,j The variance;
[0018] Step 3: Convert and generate the pixel and probability distribution of image x0;
[0019] Step 3.1: Normalize X to obtain the normalized pixel set X′={x′ i,j |i=1,2,…,a;j=1,2,…,b;where x′ i,j This represents the normalized pixel in the i-th row and j-th column;
[0020] Step 3.2: Use equations (1) and (2) to analyze Y. i,j Perform translation and scaling to obtain the mean. and variance Thus, the transformed distribution is obtained. This leads to the transformed distribution set Y':
[0021]
[0022]
[0023] In equations (1) and (2), A and B represent coefficients and constants, respectively;
[0024] Step 4: Message embedding;
[0025] Step 4.1: Calculate Y′ using equation (3). i,j probability density function f i,j (x) in the interval [x′ i,j +k-0.5,x′ i,j Integrating [+k+0.5] yields the pixel value x′. i,j The probability π under the modification magnitude k i,j (k):
[0026]
[0027] In equation (3), k represents the magnitude of the modification;
[0028] Step 4.2: Following step 4.1, obtain the modification probability {π} of each pixel X′ in X′. i,j (k)|i=1,2,…,a;j=1,2,…,b} and construct the modification pattern e, thereby obtaining the encrypted image x′0 using the STC encoding shown in equation (4):
[0029]
[0030] In equation (4), H is the parity check matrix.
[0031] The color natural image steganography method based on the diffusion model described in this invention is also characterized in that the G1 algorithm in step 2.2 obtains the pixel set X and the pixel distribution set Y according to the following process:
[0032] Based on the diffusion model, x T During processing, the output image {x} starts from time T and continues until time t=0. t The pixel in the i-th row and j-th column of |t∈{T,T-1,···,0} mean and variance Where, x t This represents the image output by the diffusion model at time t. x represents t The pixel in the i-th row and j-th column, x represents t The pixel in the i-th row and j-th column The mean, x represents t The pixel in the i-th row and j-th column The variance is obtained by using equation (6) to obtain the pixel x in the i-th row and j-th column of the image x0 output by the diffusion model at time 0. i,j :
[0033]
[0034] In equation (6), z represents a sampled value randomly sampled from the standard normal distribution; The graph x represents the time t=1. t The pixel in the i-th row and j-th column The mean, The graph x represents the time t=1. t The pixel in the i-th row and j-th column The variance.
[0035] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the color image steganography method, and the processor is configured to execute the program stored in the memory.
[0036] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the color image steganography method.
[0037] Compared with existing technologies, the beneficial effects of this invention are reflected in:
[0038] 1. This invention proposes an information hiding method based on a diffusion model, which not only does not affect the generation capability of the original diffusion model, but also achieves the ability to securely transmit secret information.
[0039] 2. The encrypted image in this invention has a larger size and higher image quality. Existing solutions mostly target grayscale images and small-sized images; in comparison, this invention is more in line with real-world usage scenarios, making the method more secure.
[0040] 3. This invention uses the generation probability of a diffusion model to embed secret information, not only embedding in a part of the image region, but considering the embedding of the entire image region, which greatly improves the embedding capacity.
[0041] 4. Modified steganography suffers from low security because embedding messages alters the probability distribution of the original image. In this invention, the probability distribution of image pixels is calculated, and the message embedding process does not change the distribution parameters. This means the difference between the steganographic image and the normally generated image is minimal, giving the steganographic image high resistance to steganography analysis.
[0042] 5. This invention provides a method for calculating and obtaining the state change distribution during the generation process of a diffusion model, so as to develop new steganography schemes based on this.
[0043] 6. In the application of this invention, the communicating parties can transmit secret messages over a public channel using a publicly available diffusion model. As an emerging model, the diffusion model has a strong ability to generate high-quality and large-size samples. By leveraging the model's generation process, the payload, quality, size, and security of the steganography method can be improved. This allows full utilization of the diffusion model's good generation effect and the calculation of pixel probabilities during the generation process, thereby increasing the quality of the steganography image while simultaneously increasing the amount of secret information carried. Attached Figure Description
[0044] Figure 1 This is a schematic diagram of the entire invention;
[0045] Figure 2 This is a schematic diagram illustrating the operation steps of the present invention. Detailed Implementation
[0046] In this embodiment, a color image steganography method based on a diffusion model is applied to, for example... Figure 1 The diagram illustrates the transmission of confidential information over public channels to prevent leakage and theft, thus achieving covert communication. Figure 2 Here is the overall flowchart of the method, which proceeds in the following steps:
[0047] Step 1: Preprocessing of Confidential Information
[0048] Step 1.1: Encrypt the confidential information to be hidden using an encryption mechanism to obtain an encrypted sequence;
[0049] Step 1.2: Use a converter to convert the encrypted sequence into a binary sequence m;
[0050] Step 2: Use the generation process of the diffusion model to obtain the generated image pixels and probability distribution;
[0051] Step 2.1, from the normal distribution Randomly sample a×b×c values to form a set W, and combine the sampled values in set W according to the size a×b×c to form the input matrix x. TWhere a and b are the height and width of the image, and c represents the number of channels in the image;
[0052] Step 2.2: Input matrix x T The input diffusion model is used to obtain the generated image x0 output by the diffusion model, and the algorithm G1 is used to process x0. T The process involves obtaining each pixel of the generated image x0 and its pixel distribution, and then denoting the set of pixel values composed of the pixel values of each pixel in the generated image x0 as X = {x}. i,j |i=1,2,…,a;j=1,2,…,b}, where x i,j This represents the pixel value in the i-th row and j-th column; it generates the pixel value x in the i-th row and j-th column of the image x0. i,j The distribution is denoted as in, x represents i,j The mean, x represents i,j The variance; let the set of pixel value distributions of X be denoted as Y = {Y i,j |i=1,2,…,a;j=1,2,…,b};~ indicates obedience.
[0053] Specifically, G1 calculates the probability and probability distribution of the generated image, including:
[0054] Step A: The diffusion model uses the input matrix x T As input, the image x at time t is output through the diffusion model of equation (5). t Each pixel mean and variance
[0055]
[0056] Where f(·) represents the diffusion network; t represents different moments in the iterative output of the diffusion model, t∈{T,T-1,…,0}; and T is the total iteration time.
[0057] Step B: Use equation (6) to obtain the pixel x of x0. i,j Pixel values:
[0058]
[0059] Where z represents a sampled value randomly sampled from a standard normal distribution.
[0060] Step 3: As Figure 2 As shown, after obtaining the generated image pixels and probability distribution, a transformation is required to obtain new image pixels and probability distribution;
[0061] Step 3.1: Normalize the set X to obtain the normalized set denoted as X′={x′}. i,j |i=1,2,…,a; j=1,2,…,b};
[0062] Step 3.2: Use equations (1) and (2) to analyze Y. i,j Perform translation and scaling transformations, and calculate the mean. and variance Obtain the transformed distribution Thus, the transformed distribution set Y' is obtained:
[0063]
[0064]
[0065] In equations (1) and (2), A and B are the coefficients and constants for normalizing set X, respectively.
[0066] Step Four: As Figure 2 As shown, the modified probability and modification pattern are calculated using the transformed image pixels and probability distribution to complete message embedding;
[0067] Step 4.1: Calculate f using equation (3) i,j (x) in the interval [x′ i,j +k-0.5,x′ i,j The integral of [+k+0.5] yields π. i,j (k) represents pixel x' i,j The probability of modifying the magnitude k.
[0068]
[0069] In equation (3), k is the change in the magnitude of the modification of the pixel, and f i,j (x) represents the distribution Y′ in set Y′. i,j The probability density function.
[0070] Step 4.2: According to the modification probability π of each pixel in Step 4.1 i,j (k), obtain the modified mode e, and then obtain the secret image x′0 through STC encoding, as shown in equation (4):
[0071]
[0072] In equation (4), H is the parity check matrix.
[0073] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.
[0074] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.
Claims
1. A color image steganography method based on diffusion model, characterized in that, The procedure is as follows: Step 1: Preprocessing of confidential information; Step 1.1: After encrypting the confidential information using an encryption mechanism, an encrypted sequence is obtained; Step 1.2: Use a converter to convert the encrypted sequence into a binary sequence. ; Step 2: Obtain the pixel count and probability distribution of the generated image; Step 2.1, from the normal distribution Randomly sample a×b×c values to construct the input image. Where a and b are the height and width of the input image, and c is the number of channels of the input image; Step 2.2: Input image The input is processed in a diffusion model, and the generated image is output. ; Using the G1 algorithm Processing is performed to obtain Each pixel and its distribution in the generated image. The set of pixels composed of all pixels in the image is denoted as . ,in, express The pixel in the i-th row and j-th column; The set of pixel distributions is denoted as ; Represents the pixel in the i-th row and j-th column. The distribution, and ;in, To show obedience, express The mean, express The variance; Step 3: Convert and generate an image The pixel and probability distribution; Step 3.1, for Normalization is performed to obtain the normalized set of pixels. ;in, This represents the normalized pixel in the i-th row and j-th column; Step 3.2: Use equations (1) and (2) to... Perform translation and scaling to obtain the mean. and variance Thus, the transformed distribution is obtained. This leads to the transformed distribution set. ': (1) (2) In equations (1) and (2), A and B represent coefficients and constants, respectively; Step 4: Message embedding; Step 4.1: Calculate using equation (3) probability density function The pixel value is obtained by integration. Probability under modification magnitude k : (3) In equation (3), k represents the magnitude of the modification; Step 4.2, obtain according to step 4.1 Each pixel Modification probability And form a modified mode e, thereby using the STC encoding shown in equation (4) to obtain the encrypted image. : (4) In equation (4), H is the parity check matrix.
2. The color image steganography method based on a diffusion model according to claim 1, characterized in that, The G1 algorithm in step 2.2 obtains the pixel set as follows: and pixel distribution set : Based on the diffusion model During processing, the images are output from time T until time t=0. The pixel in the i-th row and j-th column mean and variance ; in, This represents the image output by the diffusion model at time t. express The pixel in the i-th row and j-th column, express The pixel in the i-th row and j-th column The mean, express The pixel in the i-th row and j-th column The variance is used to obtain the image output by the diffusion model at time 0 using equation (6). The pixel in the i-th row and j-th column : (6) In equation (6), This represents a sample value randomly sampled from a standard normal distribution; Image representing time t=1 The pixel in the i-th row and j-th column The mean, Image representing time t=1 The pixel in the i-th row and j-th column The variance.
3. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the color image steganography method of claim 1 or 2, and the processor is configured to execute the program stored in the memory.
4. A computer-readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to perform the steps of the color image steganography method of claim 1 or 2.
Citation Information
Patent Citations
Secure non-embedded steganography method based on generative adversarial network
CN112243005A
Generative image steganography method based on diffusion model
CN116456037A
