Carrier-free information hiding method based on de-noising diffusion probability model
The high-quality steganographic images are generated through the denoising diffusion probability model, which solves the problems of low image quality and low extraction accuracy in carrier-free information hiding technology, and realizes high-security information embedding and extraction.
Patent Information
- Application Number
- CN202411004873.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-12-03
AI Technical Summary
The existing uncarrier information hiding technology generates low quality of hidden images, low accuracy of hidden information extraction during decryption, and low steganography capacity.
Using the denoising diffusion probability model, the denoising diffusion probability model is used to convert secret information into bitstream data and encrypt it, and the denoising diffusion implicit sampling algorithm is used to generate high-quality steganography images to achieve no The carrier information is hidden.
It improves the image quality and secret information embedding capacity of the carrier image, enhances the accuracy of ciphertext information extraction, and can avoid the detection of most steganography analysis tools, which is highly secure.
Smart Images

Figure CN119071402A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of carrier-free information hiding, and more specifically, to a carrier-free information hiding method based on a denoising diffusion probability model. Background Art
[0002] The goal of steganography is to embed secret information in an inconspicuous carrier that the sender can transmit to the receiver through an open channel without any malicious third party being able to detect the existence of the secret information.
[0003] Images are the perfect choice for steganography carriers. Traditional image steganography schemes embed secret information into carrier images by modifying the carrier image pixels or coefficients in the transform domain, which will modify the carrier image. Therefore, traditional image steganography schemes will inevitably leave traces of modification, which will be detected by steganalysis algorithms. In order to overcome the shortcomings of traditional image steganography schemes, carrier-free information hiding technology has been proposed. Carrier-free information hiding does not mean that a carrier image is not needed, but that the carrier image is synthesized by secret information. The synthesized carrier image can be regarded as the sum of the secret information. The carrier image used in the carrier-free information hiding scheme is generated by synthesis and has not been modified. Therefore, it will not be detected by traditional steganalysis algorithms and has higher security.
[0004] At present, it is proposed to apply the diffusion model to carrierless information hiding. The processed secret information is input into the diffusion model to generate a secret image. In the decryption process, the corresponding secret information is obtained by inputting the secret image into the extraction network. However, the secret image directly generated by the diffusion model has low image quality. In the decryption process, there are problems such as low accuracy of hidden information extraction and low steganographic capacity. The denoising diffusion probability model is a new high-quality image synthesis technology that can transform a noisy image into a synthetic image similar to the training data set, but the process is irreversible, so it cannot be directly applied to carrierless information hiding. Summary of the invention
[0005] In order to overcome the defects of the above-mentioned existing carrier-free information hiding technology, such as low quality of the generated secret image, low accuracy of hidden information extraction and low steganographic capacity in the decryption process, the present invention provides a carrier-free information hiding method based on a denoising diffusion probability model.
[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:
[0007] A carrier-free information hiding method based on a denoising diffusion probability model comprises the following steps:
[0008] After converting the secret information into bit stream data, it is encrypted using a stream cipher to obtain encrypted data;
[0009] According to a preset mapping rule, the encrypted data is mapped to generate a noise image that approximately obeys a standard normal distribution;
[0010] The noise image is input into a trained denoising diffusion probability model, and a steganographic image is obtained by using a denoising diffusion implicit sampling algorithm.
[0011] Furthermore, the present invention also proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, the processor executes the steps of the carrier-free information hiding method as described in the present invention.
[0012] Furthermore, the present invention also proposes a storage medium on which computer-readable instructions are stored, wherein the computer-readable instructions, when executed by a processor, implement the steps of the carrier-free information hiding method as described in the present invention.
[0013] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:
[0014] The present invention uses the inverse algorithm of the denoising diffusion probability model sampling algorithm to complete carrier-free information hiding, while improving the image quality of the carrier image, so that the carrier image has a higher secret information embedding capacity and ciphertext information extraction accuracy; the present invention generates a carrier image by driving the secret information, and compared with artificially modifying the pixel value of the carrier image, it can evade the detection of most steganalysis tools and has the characteristic of high security. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 This is a flow chart of the carrier-free information hiding method based on the denoising diffusion probability model of Example 1.
[0016] Figure 2 Schematic diagram of the training process of the denoising diffusion probability model of Example 1.
[0017] Figure 3 This is a flowchart of the carrier-free information hiding method based on the denoising diffusion probability model of Example 2. DETAILED DESCRIPTION
[0018] The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present invention;
[0019] It is understandable to those skilled in the art that some well-known descriptions may be omitted in the drawings.
[0020] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0021] Example 1
[0022] This embodiment proposes a carrier-free information hiding method based on a denoising diffusion probability model. Figure 1 , which is a flow chart of the carrier-free information hiding method based on the denoising diffusion probability model of this embodiment.
[0023] The carrier-free information hiding method based on the denoising diffusion probability model proposed in this embodiment includes the following steps:
[0024] S1, after converting the secret information into bit stream data, encrypt it using a stream cipher to obtain encrypted data;
[0025] S2. According to a preset mapping rule, the encrypted data is mapped to generate a noise image that approximately obeys a standard normal distribution;
[0026] S3. Input the noise image into a trained denoising diffusion probability model, and obtain a stego image using a denoising diffusion implicit sampling algorithm.
[0027] This embodiment proposes to apply the inverse algorithm of the denoising diffusion probability model sampling algorithm, namely the denoising diffusion implicit sampling algorithm, to realize the conversion of noisy image-synthetic image-noise image and complete the carrier-free information hiding. At the same time, the image quality of the carrier image is improved by applying the denoising diffusion probability model, so that the carrier image has a higher secret information embedding capacity and ciphertext information extraction accuracy.
[0028] Among them, the secret information steganist converts the secret information into binary bit stream data and encrypts the secret information using a stream cipher. Then, the encrypted secret information is mapped into a noise image that approximately obeys the standard normal distribution, and then the noise image is input into the denoising diffusion probability model, and a carrier image is generated using the forward sampling algorithm to complete the carrier-free information hiding.
[0029] Therefore, the carrier image generated by this embodiment is driven by secret information, instead of manually modifying the pixel value of the carrier image, so it can evade detection by most steganalysis tools and has the characteristic of high security.
[0030] In addition, this embodiment establishes a mapping relationship between secret information and noise image through a preset mapping rule, and does not require an additional secret information extraction model. Optionally, after receiving the stego image, the secret information receiver can restore the noise image through inverse sampling, and obtain the corresponding encrypted data according to the inverse mapping of the mapping rule.
[0031] In an optional embodiment, in step S1, the converting of the secret information into bit stream data and then encrypting it using a stream cipher comprises the following steps:
[0032] S1.1. Generate a binary key stream of the same length as the secret information using a key generation algorithm;
[0033] S1.2. After converting the secret information into bit stream data, perform bit-wise XOR operation on the secret information and the binary key stream to obtain binary encrypted data.
[0034] As an exemplary description, this embodiment selects the RC4 key generation algorithm to generate the key stream.
[0035] In an optional embodiment, in step S2, mapping the encrypted data to generate a noise image according to a preset mapping rule includes the following steps:
[0036] S2.1, preprocessing the encrypted data, for encrypted data whose information length is less than row×col×channel, alternately padding 0 and 1 until the length is row×col×channel; wherein row is the length of the noise image, col is the width of the noise image, and channel is the number of channels;
[0037] S2.2, traverse the preprocessed encrypted data, for any bit B i , if its bit value is 0, add its serial number i to the Zeros list, otherwise add its serial number i to the Ones list;
[0038] S2.3, using a random number generator to generate a random number sequence R with a length consistent with the encrypted data and uniformly distributed, where R i ∈{0,1};
[0039] S2.4, traverse the random number sequence R, for any random number R i , according to whether its subscript i belongs to the Zeros list or the Ones list, the random number R i Transform according to the preset rules, where the random number R whose subscript i belongs to the Zeros list i Convert to a value less than 0, and convert the random number R whose index i belongs to the list of Ones i Converted to a value greater than 0, and the deformed random number sequence R approximately obeys the standard normal distribution;
[0040] S2.5. Map the deformed random number sequence R into a matrix of row×col×channel size to obtain a noise image X, which can be approximately regarded as obeying a standard normal distribution.
[0041] In this step, the binary encrypted data bit stream is complemented based on the size of the noise image, so as to facilitate mapping the encrypted data to the noise image. The binary encrypted data bit stream can be approximately regarded as uniformly distributed.
[0042] In an optional embodiment, step S2.2 further includes the following steps: dividing the Zeros list and the Ones list into Zeros and Ones according to a preset ratio. 1 、Zeros 2 and Zeros 3 Subsets, and Ones 1 、Ones 2 and Ones 3 Subset;
[0043] In step S2.4, the random number R is selected according to whether its subscript i belongs to the Zeros list or the Ones list. i Transform according to the preset rules, including: if the subscript i∈Zeros 1 , then let the random number R i =R i -1; if i∈Zeros 2 , then let the random number R i =R i -2; if i∈Zeros 3 , then let the random number R i =R i -3; if i∈Ones 1 , then the random number R i unchanged; if i∈Ones 2 , then let the random number R i =R i +1; if i∈Ones 3 , then let the random number R i =R i +2. Its expression is:
[0044]
[0045] Through this transformation rule, the bit value "0" in the secret information in this embodiment is converted into data less than 0 in the noise image, and the bit value "1" is converted into data greater than 1 in the noise image. When the secret information is restored, the mapping rule can be reversed to traverse the pixel values of the noise image, convert the pixel values less than 0 into the bit value "0", and convert the pixel values greater than 0 into the bit value "1", so that the encrypted data is restored.
[0046] In this embodiment, by traversing the binary encrypted data bit stream, the element numbers corresponding to the bit values of 1 and 0 are added to the corresponding list, and the list is divided into 6 sets according to a preset ratio, and elements are added by random selection when dividing the subsets. The subsets thus divided are combined with the deformation rules of random numbers to obtain a noise image that approximately obeys the standard normal distribution.
[0047] It should be noted that the property of data that obeys the normal distribution is that the range of values within one standard deviation from the mean accounts for 68% of all values, the ratio within two standard deviations accounts for 95%, and the ratio within three standard deviations accounts for 99%. Therefore, to obtain a data set that obeys the normal distribution, we must first divide the subsets according to this characteristic.
[0048] Further optionally, in step S2.2, when the Zeros list and the Ones list are divided according to a preset ratio, the preset ratio is: Zeros 1 Subsets, Ones 1 The subsets contain 68% to 81% of the elements in the original list, corresponding to data that is less than one standard deviation from the mean; Zeros 2 Subsets, Ones 2 The subsets contain 27% to 14% of the elements in the original list, and the subsets correspond to data that are greater than one standard deviation and less than two standard deviations from the mean; Zeros 3 Subsets, Ones 3 The subsets contain the remaining 5% of the elements in the original list, and the subsets correspond to the data that are greater than two standard deviations and less than three standard deviations from the mean. When dividing the subsets, the elements are added by random selection.
[0049] Further optionally, in step S2.2, when the Zeros list and the Ones list are divided according to a preset ratio, the Zeros list and the Ones list are divided according to a preset ratio. 1 Subsets, Ones 1 The subsets contain 81% of the elements in the original list, Zeros 2 Subsets, Ones 2 The subsets each contain 14% of the elements in the original list, Zeros 3 Subsets, Ones 3 The subsets each contain the remaining 5% of the elements in the original list.
[0050] Since the encrypted bit stream does not strictly follow a uniform distribution, in experiments we found that a first subset of elements with a proportion of 81% can achieve better results.
[0051] Furthermore, to obtain a noise image, a mapping operation must be performed again, and the mapping operation must be reversible. Therefore, a deformation rule can be set to map the corresponding secret information so that the noise image can be approximately regarded as obeying a standard normal distribution. This embodiment generates a set of random number sequences with values of 0 or 1, and determines whether the random number sequence belongs to Zeros. 1 、Zeros 2 、Zeros 3 、Ones 1 、Ones 2 or Ones 3 Which subset of the random number sequence is used to execute the corresponding deformation rule to obtain a random number sequence that is deformed and approximately obeys the standard normal distribution, and then further mapped into a noise image of row×col×channel. When recovering the secret information, the deformation rule can be reversed to restore the binary encrypted data bit stream by traversing the pixel values of the noise image.
[0052] In an optional embodiment, in step S3, the noise image is input into a trained denoising diffusion probability model, and a steganographic image is obtained by using a denoising diffusion implicit sampling algorithm, including the following steps:
[0053] Input the noisy image X into the trained denoising diffusion probability model to obtain the noise prediction value
[0054] For the noisy image x T Using the denoising diffusion implicit sampling algorithm, the stego image x′ is obtained 0 ; Its expression is:
[0055]
[0056] Where t is the acceleration time step, and t∈{0,s,2s,...,Ts,T}, T is the preset total step length, and s is the preset sampling interval; when t=T, let x′ T =x T ; In each iterative calculation process, the image x′ t The noise prediction value output after inputting the denoising diffusion probability model; is a set of linearly increasing hyperparameters, β t =β 0 +(β T -β 0 )t,β 0 and β T is the preset hyperparameter. Through iterative processing, the stego-image x′ is obtained. 0 .
[0057] In this embodiment, a denoising diffusion implicit sampling algorithm is applied to generate a high-quality image from random noise in a step-by-step denoising manner.
[0058] Exemplarily, the maximum time step T is set to 1000; if sampling is required with a total number of steps of 200, then t∈{0,5,10,15,...,990,995,1000}, that is, the sampling interval s=5.
[0059] Further optionally, the denoising diffusion probability model includes a backbone network of a U-Net structure. The method further includes the following steps:
[0060] Get a training image set;
[0061] For any training image x 0 , add T steps of Gaussian noise that obeys the standard normal distribution, and get the corresponding noisy training image x 1 ,x 2 ,…,x n …,x T ; Its expression is:
[0062]
[0063] Where n∈[0,T] and n∈Z; is a set of linearly increasing hyperparameters, ∈ is a noise vector that follows a standard normal distribution and has the same size as the input image;
[0064] Add the noise to the training image x n and the corresponding time step n is input into the denoising diffusion probability model, and the denoising diffusion probability model outputs a noise prediction value ∈′;
[0065] A loss function is constructed based on the mean square error between the noise prediction value ∈′ and the true value ∈, and the denoising diffusion probability model is optimized and trained until the model parameters converge or reach a preset cutoff condition, thereby obtaining a denoising diffusion probability model that has completed training.
[0066] like Figure 2 , which is a schematic diagram of the training process of the denoising diffusion probability model of this embodiment.
[0067] In the training process of the denoising diffusion probability model, when x is known n-1 In the case of adding Gaussian noise, the noisy training image x is obtained. n The expression is:
[0068]
[0069] In the formula, β n =β 0 +(βT -β 0 )n,β 0 and β T is the preset hyperparameter; the process of adding noise is the forward noise adding process, which gradually adds noise to the original distribution and converts it into a priori distribution Obviously, the noise addition process is a Markov chain process, because the output of each time step t depends on the output x of the previous time step n-1 .
[0070] Based on the standard normal distribution and the special properties of the Markov chain, the noisy training image x n The above formula is converted to:
[0071] Through the above formula, as long as a certain time step n is given, the noise-added training image x corresponding to the time step can be directly calculated. n , it is no longer necessary to follow the chain process calculation. Furthermore, due to is a constant. It can be inferred from the above formula that the noisy image x at a certain time step n is known. n , and the noise ∈ added at this time step, the input image x can be restored 0 , denoising is completed.
[0072] This embodiment trains the denoising diffusion probability model, and its training goal is to give a certain time step n and the corresponding noisy image x n , the model is able to predict the added noise ∈.
[0073] For example, the anime avatar dataset and the CelebA celebrity dataset are used as training image sets. The anime avatar dataset contains 21,548 images of size 64×64, and the CelebA dataset contains 202,599 images of size 64×64.
[0074] For example, we set the hyperparameter β 0 =0.0001, β T =0.02.
[0075] Exemplarily, the denoising diffusion probability model is optimized and trained using an AdamW optimizer.
[0076] In this embodiment, by applying the denoising diffusion probability model to the field of carrier-free information hiding, the image quality of the carrier image is greatly improved, and it also has a higher secret information embedding capacity and ciphertext information extraction accuracy.
[0077] Example 2
[0078] This embodiment proposes a carrier-free information hiding method based on a denoising diffusion probability model, comprising the following steps:
[0079] S101, converting secret information into bit stream data, and encrypting it using a stream cipher to obtain encrypted data;
[0080] S102, mapping the encrypted data to generate a noise image that approximately obeys a standard normal distribution according to a preset mapping rule;
[0081] S103, inputting the noise image into a trained denoising diffusion probability model, and obtaining a stego image using a denoising diffusion implicit sampling algorithm;
[0082] transmitting the stego-image to a secret receiver via a public channel;
[0083] S201, after receiving the stego image, the secret receiver performs inverse sampling based on a denoising diffusion implicit sampling algorithm on the stego image to obtain a noise image;
[0084] S202, extracting the encrypted data according to the inverse process of the mapping rule;
[0085] S203: Decrypt the encrypted data using the same stream cipher to obtain the secret information.
[0086] This embodiment applies the carrier-free information hiding method proposed in Embodiment 1 and proposes a restoration process of the steganographic image by the secret receiver. Figure 3 FIG. 1 is a flowchart of the carrier-free information hiding method of this embodiment.
[0087] The stego image is obtained by applying the denoising diffusion implicit sampling algorithm, so the received stego image x′ 0 By performing the inverse sampling operation of the denoising diffusion implicit sampling algorithm, the noisy image X can be restored.
[0088] In an optional embodiment, performing inverse sampling based on a denoising diffusion implicit sampling algorithm on the stego-image comprises the following steps:
[0089] Based on the inverse sampling formula of the denoising diffusion implicit sampling algorithm, the stego image x′ 0 Restored to the noisy image X; its expression is:
[0090]
[0091] where t∈{0,s,2s,...,Ts,T}; the image x′ is obtained after the inverse sampling iterative processing T That is the noise image X.
[0092] Furthermore, the encrypted data is obtained by extracting the restored noise image X according to the inverse process of the mapping rule.
[0093] In an optional embodiment, extracting the encrypted data according to the inverse process of the mapping rule includes the following steps:
[0094] Initialize the bit sequence S to be an empty sequence;
[0095] Traverse each pixel in the noisy image restored by the inverse sampling algorithm: if the pixel value is greater than 0, add 1 to the bit sequence S, otherwise add 0; after completing the traversal, the restored encrypted data bit stream is obtained.
[0096] This embodiment traverses the pixel values of the noise image, converts pixel values less than 0 into bit values "0", and converts pixel values greater than 0 into bit values "1", thereby completing the restoration of the encrypted data.
[0097] Furthermore, the same stream cipher can be used to obtain the bitstream secret information embedded by the secret steganologist.
[0098] This embodiment establishes a mapping relationship between secret information and noise images through preset mapping rules. After receiving the steganographic image, the secret information receiver can restore the noise image through inverse sampling, and can obtain the corresponding encrypted data according to the inverse mapping of the mapping rules. No additional secret information extraction model is required, which effectively reduces the cost of computing resources.
[0099] Example 3
[0100] This embodiment proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the carrier-free information hiding method proposed in Embodiment 1 or Embodiment 2.
[0101] Example 4
[0102] This embodiment proposes a storage medium on which computer-readable instructions are stored, wherein the computer-readable instructions, when executed by a processor, implement the steps of the carrier-free information hiding method proposed in Embodiment 1 or Embodiment 2.
[0103] Exemplarily, the storage medium includes, but is not limited to, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0104] Exemplarily, the instructions, programs, code sets or instruction sets may be implemented using conventional programming languages.
[0105] Exemplarily, the processor includes but is not limited to a smart phone, a personal computer, a server, a network device, etc., and is used to execute all or part of the steps of the carrier-free information hiding method proposed in Example 1 or Example 2.
[0106] The terms in the drawings are only used for illustrative purposes and are not to be construed as limiting the present invention;
[0107] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.
Claims
1. A carrier-free information hiding method based on a denoising diffusion probability model, characterized in that: The following steps are involved: After converting the secret information into bit stream data, it is encrypted using a stream cipher to obtain encrypted data; According to a preset mapping rule, the encrypted data is mapped to generate a noise image that approximately obeys a standard normal distribution; The noise image is input into a trained denoising diffusion probability model, and a steganographic image is obtained by using a denoising diffusion implicit sampling algorithm.
2. The carrier-free information hiding method according to claim 1, characterized in that: After the secret information is converted into bit stream data, it is encrypted using a stream cipher, comprising the following steps: Using a key generation algorithm, a binary key stream having the same length as the secret information is generated; After the secret information is converted into bit stream data, the secret information is XORed with the binary key stream on a bit-by-bit basis to obtain binary encrypted data.
3. The carrier-free information hiding method according to claim 1, characterized in that: The step of mapping the encrypted data to generate a noise image according to a preset mapping rule comprises the following steps: Preprocess the encrypted data, and for encrypted data whose length is less than row×col×channel, alternately fill it with 0 and 1 until the length is row×col×channel; wherein row is the length of the noise image, col is the width of the noise image, and channel is the number of channels; Traverse the preprocessed encrypted data, for any bit B i , if its bit value is 0, add its serial number i to the Zeros list, otherwise add its serial number i to the Ones list; A random number generator is used to generate a random number sequence R whose length is consistent with the encrypted data and obeys a uniform distribution, where R i ∈{0,1}; Traverse the random number sequence R, for any random number R i , according to whether its subscript i belongs to the Zeros list or the Ones list, the random number R i Transform according to the preset rules, where the random number R whose subscript i belongs to the Zeros list i Convert to a value less than 0, and convert the random number R whose index i belongs to the list of Ones i Converted to a value greater than 0, and the deformed random number sequence R approximately obeys the standard normal distribution; The deformed random number sequence R is mapped into a matrix of size row×col×channel to obtain the noisy image X.
4. The carrier-free information hiding method according to claim 3, characterized in that: The step of mapping the encrypted data to generate a noise image according to a preset mapping rule further includes the following steps: Dividing the Zeros list and the Ones list into Zeros1, Zeros2 and Zeros3 subsets, and Ones1, Ones2 and Ones3 subsets respectively according to a preset ratio; The random number R is obtained according to whether its subscript i belongs to the Zeros list or the Ones list. i Transform according to the preset rules, including: if the subscript i∈Zeros1, then let the random number R i =R i -1; if i∈Zeros2, let the random number R i =R i -2; if i∈Zeros3, let the random number R i =R i -3; if i∈Ones1, then the random number R i unchanged; if i∈Ones2, let the random number R i =R i +1; if i∈Ones3, let the random number R i =R i +2.
5. The carrier-free information hiding method according to any one of claims 1 to 4, characterized in that: The step of inputting the noise image into a trained denoising diffusion probability model and obtaining a stego image using a denoising diffusion implicit sampling algorithm comprises the following steps: Input the noisy image X into the trained denoising diffusion probability model to obtain the noise prediction value For the noisy image x T The steganographic image x′0 is obtained by using the denoising diffusion implicit sampling algorithm; its expression is: Where t is the acceleration time step, and t∈{0,s,2s,...,Ts,T}, T is the preset total step length, and s is the preset sampling interval; when t=T, let In each iterative calculation process, the image x′ t The noise prediction value output after inputting the denoising diffusion probability model; is a set of linearly increasing hyperparameters; through iterative processing, the stego-image x′0 is obtained.
6. The carrier-free information hiding method according to claim 5, characterized in that: The denoising diffusion probability model includes a backbone network of a U-Net structure; the method further includes the following steps: Get a training image set; For any training image x0, add T steps of Gaussian noise that follows the standard normal distribution to obtain the corresponding noisy training images x1, x2, …, x n …,x T ; Its expression is: Where n∈[0,T] and n∈Z; is a set of linearly increasing hyperparameters, v is a noise vector that follows a standard normal distribution and has the same size as the input image; Add the noise to the training image x n and the corresponding time step n is input into the denoising diffusion probability model, and the denoising diffusion probability model outputs a noise prediction value ∈′; A loss function is constructed based on the mean square error between the noise prediction value ∈′ and the true value ∈, and the denoising diffusion probability model is optimized and trained until the model parameters converge or reach a preset cutoff condition, thereby obtaining a denoising diffusion probability model that has completed training.
7. The carrier-free information hiding method according to claim 5, characterized in that: The method further comprises the following steps: Performing inverse sampling based on a denoising diffusion implicit sampling algorithm on the stego-image to obtain a noise image; Extracting the encrypted data according to the inverse process of the mapping rule; The encrypted data is decrypted using the same stream cipher to obtain the secret information.
8. The carrier-free information hiding method according to claim 7, characterized in that: The performing of inverse sampling based on a denoising diffusion implicit sampling algorithm on the stego-image comprises the following steps: Based on the inverse sampling formula of the denoising diffusion implicit sampling algorithm, the stego image x′0 is restored to the noise image X; it is expressed as: where t∈{0,s,2s,...,Ts,T}; the image x′ is obtained after the inverse sampling iterative processing T That is the noise image X.
9. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by the processor, the processor is caused to perform the steps of the carrier-less information hiding method according to any one of claims 1 to 8.
10. A storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, the steps of the carrier-free information hiding method according to any one of claims 1 to 8 are implemented.
Citation Information
Cited By
Carrier image content consistency enhancement method for high-capacity steganography
CN120259133A
Image steganography method, program product, storage medium and electronic equipment
CN120856834A
