Batch image steganography method suitable for social network and related equipment

Abstract

The invention provides a batch image steganography method suitable for a social network and related equipment. According to the method, high-dimensional steganography analysis features are utilized, and an embedded trace function is designed to describe steganography distortion through high-separability feature selection and difference measurement, so that the safety of a steganography detection algorithm is enhanced. Then, a carrier selection algorithm based on embedded traces and steganography fitness and an effective load distribution strategy based on a multi-stage optimal decision are provided, so that load distribution with the minimum embedded traces is achieved, and the method can be combined with any image steganography algorithm and universal joint information source / channel coding, and has the advantages of being high in robustness and high in robustness. And a high-capacity, high-robustness and high-concealment secret-carrying image is generated.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular to a batch image steganography method and related equipment suitable for social networks. Background Art

[0002] As one of the cutting-edge technologies for protecting information security, image steganography can achieve safe, reliable and covert transmission of secret information, and has attracted extensive attention from researchers in recent years. Generally speaking, the security of image covert communication is guaranteed by generating stego images that conform to statistical rules and communication behaviors that conform to common sense. Under this premise, online social networks with massive complex and diverse images and a large user base can provide a natural and ideal environment for image covert communication.

[0003] Figure 1 A typical image steganography model suitable for online social networks is given. After the sender performs source / channel coding on the message, the secret message is first embedded into several natural images using an image steganography algorithm. Then, these images are uploaded to widely used online social networks such as Facebook, Twitter, Weibo, etc. At the receiving end, the user can obtain the secret message from the downloaded image and decode it using the shared parameters, thereby achieving covert communication that takes into account both content security and behavior security. By analyzing the above propagation scenarios, we can find that the channel provided by online social networks for image steganography has the following characteristics: (1) The images shared in online social networks are mainly color images in JPEG format; (2) The sender tends to upload multiple content-related natural images each time; (3) Portable devices with limited computing power are usually used to publish shared images; (4) Shared images carrying secret information may be subject to lossy processing during transmission. Therefore, it is necessary to design a lightweight and robust batch JPEG image steganography algorithm to meet the requirements of image covert communication in social networks.

[0004] Existing image robust steganography methods can be divided into modification-based, matching-based and generation-based algorithms. By adjusting the relationship between DCT coefficients, researchers have designed a series of image robust steganography algorithms based on the framework of "robust domain construction + minimization of embedding cost code". Although these modification-based embedding algorithms are highly robust to specific or multiple image processing attacks, they will inevitably cause distortion to the carrier image, resulting in the leakage of stego traces. To this end, researchers have proposed carrier-free image robust steganography algorithms based on multi-target recognition, hash generation models and image datasets. Since these algorithms mainly rely on the matching relationship between secret messages and image content to achieve covert communication, their embedding capacity is often quite limited and it is difficult to meet the needs of efficient communication. Recently, researchers have proposed generative image robust steganography algorithms based on flow models, gradient descent approximation and deep generative models to obtain high-quality secret images and large embedding capacity. However, these algorithms usually consume a lot of computing resources, and due to the poor reversibility of the generator, the integrity of message extraction cannot be guaranteed, which limits their application in online social networks. Summary of the invention

[0005] In order to overcome the contradiction between security, payload and robustness, the present invention proposes a batch image steganography method and related equipment suitable for social networks, aiming to minimize embedding traces, optimize payload distribution, and realize efficient and reliable covert communication in online social networks.

[0006] In a first aspect, the present invention provides a batch image steganography method applicable to social networks, comprising:

[0007] Step 1: Obtain a candidate carrier image set, use a robust steganography algorithm to perform steganographic processing on the candidate carrier image set, and generate a corresponding steganographic image set;

[0008] Step 2: for any candidate carrier-stamped image pair, measure the embedding trace of the robust steganography algorithm on the candidate carrier image;

[0009] Step 3: For any candidate carrier image, calculate its steganalytic fitness according to its embedding trace; the smaller the average embedding trace value of a candidate carrier image under different payloads, the higher the corresponding steganalytic fitness value;

[0010] Step 4: Select images with higher steganographic fitness values ​​from the candidate carrier image set to construct a carrier image sequence;

[0011] Step 5: Encode the secret message and fragment the encoded message;

[0012] Step 6: solving the optimal payload sequence; the optimal payload sequence refers to a sequence in which the lengths of message segments to be embedded in all carrier images are arranged in the order of carrier images under the condition that the embedding traces of the carrier image sequence are minimized;

[0013] Step 7: Generate a encrypted image sequence based on a minimization embedding cost algorithm according to the carrier image sequence and the optimal payload sequence;

[0014] Step 8: Utilize social networks to publish the secret image sequence for the recipient to download and extract the secret message.

[0015] Furthermore, step 2 specifically includes:

[0016] For any candidate carrier-secret image pair, extract key features from the candidate carrier image and the corresponding secret image respectively, wherein the key features refer to features that can distinguish the candidate carrier from the corresponding secret image;

[0017] According to the key features extracted from the candidate carrier image and the corresponding steganographic image, the embedding trace of the candidate carrier image by the robust steganographic algorithm is measured as follows:

[0018]

[0019] Among them, f i c and From the candidate carrier images and the corresponding encrypted image l i s The key features extracted from c (·) and τ s (·) respectively and The average value of all characteristic components in c (·) and υ s (·) respectively and The standard deviation of all feature components in .

[0020] Furthermore, for any candidate carrier-secret image pair, key features of the candidate carrier image and the corresponding secret image are extracted respectively, specifically including:

[0021] The characteristic component f is calculated as follows *,j Divisibility of A s (f *,j );

[0022]

[0023] Among them, μ c (·) and μ s (·) are the average values ​​of the j-th dimension feature components of all candidate carriers and the corresponding encrypted images, δ c (·) and δ s (·) are the standard deviations of the j-th dimension feature components of all candidate carriers and the corresponding encrypted images;

[0024] For each feature component, if its separability value is greater than the set separability threshold, the feature component is added to the key feature candidate set; based on the current key feature candidate set, the candidate carrier image and the secret carrier image are classified, if the classification accuracy is greater than the set accuracy threshold, the feature component is added to the key feature set, otherwise the feature component is deleted from the key feature candidate set.

[0025] Furthermore, in step 3, any candidate carrier image is calculated according to the following formula: The steganalytic fitness

[0026]

[0027] in, is the candidate carrier image The embedding trace of 1 ~α h Different loads in the range, λ i is the candidate carrier image Average embedding traces under different effective loads, h represents the number of values ​​of different loads.

[0028] Furthermore, step 6 specifically includes:

[0029] Calculate each carrier image c i The maximum payload is calculated when the robust steganography algorithm is used on the carrier image c i Embed k i The embedded traces of group messages after fragmentation;

[0030] Solving the optimal payload sequence is transformed into the following optimization problem;

[0031]

[0032] in, Represented in the carrier image c i Embed k i The encrypted image after the group message is fragmented. Represented in the carrier image c i Embed k i The embedding traces after group message fragmentation, k m′Indicates the total number of message fragments, l m′ Indicates the length of a single message segment, and n indicates the total number of carrier images;

[0033] A multi-stage decision optimization is adopted to solve the optimization problem and obtain the optimal payload sequence.

[0034] Furthermore, a multi-stage decision optimization is used to solve the optimization problem, specifically including:

[0035] For the carrier image set C i ={c 1 ,c 2 ,c 3 ,…,c i}, and deduce its effective load k according to the following recursive calculation j The minimum embedding trace at And record the corresponding q value at this time as G s [C i ,k j ]; where 1≤k j ≤k m′ ;

[0036]

[0037] in, Initialize to q represents the optimal result in the current stage corresponding to the carrier image set C i The number of message fragments embedded in ;

[0038] The optimal payload sequence p = {ρ 1 ,ρ 2 ,ρ 3 ,…,ρ n}, ρ i =q×l m′ .

[0039] In a second aspect, the present invention provides a batch image steganography system suitable for social networks, comprising:

[0040] The preprocessing module is used to obtain a candidate carrier image set, perform steganographic processing on the candidate carrier image set using a robust steganographic algorithm, and generate a corresponding steganographic image set;

[0041] An embedding trace measurement module, used for measuring the embedding trace of the robust steganography algorithm on any candidate carrier-stamped image pair;

[0042] The stegographic fitness calculation module is used to calculate the stegographic fitness of any candidate carrier image according to its embedding trace; wherein the smaller the average embedding trace value of a candidate carrier image under different payloads is, the higher the corresponding stegographic fitness value is;

[0043] A carrier selection module is used to select images with higher steganographic fitness values ​​from a candidate carrier image set to construct a carrier image sequence;

[0044] A message encoding module, used to encode the secret message and fragment the encoded message;

[0045] A load distribution module is used to solve the optimal effective load sequence; the optimal effective load sequence refers to a sequence in which the lengths of message segments to be embedded in all carrier images are arranged in the order of carrier images under the condition that the embedding traces of the carrier image sequence are minimized;

[0046] An embedding module, used for generating a sequence of encrypted images based on a minimization embedding cost algorithm according to the carrier image sequence and the optimal payload sequence;

[0047] The communication module is used to publish the secret image sequence using a social network so that a receiver can download and extract the secret message.

[0048] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0049] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0050] Beneficial effects of the present invention:

[0051] (1) The present invention provides a new idea for describing the traces of steganalysis embedding, thereby establishing a steganographic algorithm design mechanism that directly resists steganalysis, and directly improving the ability to resist steganalysis. By using high-separability steganalysis feature selection and difference measurement to describe the embedding traces of steganalysis, the influence of information embedding on the high-dimensional statistical characteristics of the image can be accurately described, thereby improving the detection ability of steganalysis.

[0052] (2) The present invention defines a steganographic fitness function based on embedding traces to describe image distortion under different loads, and proposes a carrier selection algorithm, which can preferentially select images with the smallest embedding traces as carrier images to improve security.

[0053] (1) The present invention designs a payload distribution strategy based on multi-stage optimal decision making, which uses different payloads and steganalysis detection features to measure the embedding traces and distribute the secret message to the image with the smallest difference in the steganalysis features between the carrier and the secret.

[0054] In summary, the present invention can use high-dimensional feature differences to characterize embedding traces and achieve optimal payload distribution with minimal embedding traces. Therefore, the generated secret image has both strong concealment and strong robustness, thereby achieving safe, reliable and efficient covert communication in online social networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a typical image steganography model suitable for online social networks;

[0056] Figure 2 A flow chart of a batch image steganography method applicable to social networks provided by an embodiment of the present invention;

[0057] Figure 3 A schematic diagram of derivation of carrier image load distribution provided by an embodiment of the present invention;

[0058] Figure 4 A schematic diagram of the structure of a batch image steganography system applicable to social networks provided by an embodiment of the present invention;

[0059] Figure 5 The average detection accuracy of the SRNet steganalysis detection network provided by the embodiment of the present invention is: (a) 0.01, 0.03, ..., 0.09 bpnzAC; (b) 0.1, 0.2, ..., 0.5 bpnzAC; wherein green corresponds to the JCRISBE robust steganography algorithm, orange corresponds to the QTRS robust steganography algorithm, and yellow corresponds to the GMAS robust steganography algorithm;

[0060] Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] like Figure 2 As shown, an embodiment of the present invention provides a batch image steganography method applicable to social networks, comprising the following steps:

[0063] S101: Obtain candidate carrier image set o represents the number of images. The robust steganography algorithm (hereinafter referred to as the steganography function ψ) is used to perform steganographic processing on the candidate carrier image set to generate the corresponding carrier image set

[0064] It is understandable that the user can select a robust steganography algorithm that is resistant to specific attacks or multiple attacks for steganography processing as needed.

[0065] S102: for any candidate carrier-stamped image pair, measuring the embedding trace of the robust steganography algorithm on the candidate carrier image;

[0066] Specifically, in order to accurately measure the embedding traces while reducing the amount of calculation, this embodiment uses the following method to measure the embedding traces of the candidate carrier image:

[0067] For any candidate carrier-secret image pair, extract key features from the candidate carrier image and the corresponding secret image respectively, wherein the key features refer to features that can distinguish the candidate carrier from the corresponding secret image;

[0068] It can be understood that high-dimensional features can be used to capture stegographic modifications in practical applications. Taking a typical rich-model stegographic feature f with dimension d as an example, we can extract the steganalysis features f from the candidate carrier image set I and the corresponding steganalysis image set I. s Extract feature set and in In order to improve the classification accuracy and computational rate, it is necessary to simplify the above feature set, that is, to select highly separable features from it to capture the embedding traces.

[0069] Further, as an implementable method, in this embodiment, the following formula is first used to measure the separability A of the feature component: s (f *,j ); Then, for each feature component, if its separability value A s (f *,j ) is greater than the set separability threshold β a , then the characteristic component f *,j Add key feature candidate set and Based on the current candidate set of key features, the candidate carrier image and the secret image are classified. If the classification accuracy is Greater than the set accuracy threshold β c , then add the feature component to the set of key features f R , otherwise the feature component is deleted from the key feature candidate set.

[0070]

[0071] Among them, μ c (·) and μ s (·) are the average values ​​of the j-th dimension feature components of all candidate carriers and the corresponding encrypted images, δ c (·) and δ s (·) are the standard deviations of the j-th dimension feature components of all candidate carriers and the corresponding encrypted images;

[0072] In the above formula, reflects the difference in the j-th dimension feature component and indicates the dispersion between classes. reflects the change of the j-th dimension feature component and indicates the aggregation within the class. According to the principle of "dispersion between classes and aggregation within classes", A s The larger (f) is, the better the effect of distinguishing the carrier and the steganographic image is. Therefore, after classifying all features and removing irrelevant feature components whose attribute separability is lower than a certain threshold, features with similar attribute separability can be compared and selected to obtain key features that can accurately capture stegographic traces.

[0073] Assume that the selected key feature set f R The dimension is r, and a pair of candidate carrier images and encrypted images Take the following as an example, and extract the key features corresponding to the two. and Then, according to the key features extracted from the candidate carrier image and the corresponding secret image, the embedding trace of the robust steganography algorithm on the candidate carrier image is measured according to the following formula:

[0074]

[0075] Among them, f i c and From the candidate carrier images and the corresponding encrypted image The key features extracted from c (·) and τ s (·) respectively and The average value of all characteristic components in c (·) and v s (·) respectively and The standard deviation of all feature components in .

[0076] It can be understood that in formula (3), It reflects the difference in external features before and after the message is embedded, indicating the dispersion of features between classes. c (f i c )and It reflects the changes in the internal features of the image after steganography, indicating the aggregation of features of the same category. It can be clearly seen that The higher the value, the greater the inter-class feature dispersion between the carrier image and the dense image, and the smaller the intra-class feature aggregation, which is also in line with the design concept of "inter-class feature dispersion and intra-cluster feature aggregation".

[0077] S103: for any candidate carrier image, calculating its steganalysis fitness according to its embedding trace; wherein, the smaller the average embedding trace value of a candidate carrier image under different payloads, the higher the corresponding steganalysis fitness value;

[0078] Specifically, formula (3) can be used to measure a single image Using steganographic functions with specific payloads In order to fully describe The impact of image statistics under each payload, defining the steganographic fitness function for different images as follows:

[0079]

[0080] Among them, ρ is α 1 ~α h Different loads in the range, λ i is the average embedding trace for all loads, Reflects the image Changes of embedding traces under different loads, h is the number of values ​​of different loads.

[0081] In the above definition, the smaller the average embedding trace value under different payloads, that is, the smaller the influence of the steganographic function on the high-dimensional features, the image The steganalytic fitness In addition, the greater the fluctuation of the embedded traces under different loads, the more it indicates that the changes in high-dimensional features are difficult to describe the degree of leakage of steganalysis traces, which in turn indicates that the image The steganalytic fitness The higher.

[0082] S104: Selecting images with higher steganographic fitness values ​​from the candidate carrier image set to construct a carrier image sequence, thereby reducing the leakage of steganographic signs and improving concealment.

[0083] Specifically, the carrier images are sorted according to their steganographic fitness, denoted as Set the steganalysis fitness threshold β f, delete F s <β f The candidate carrier images are obtained, thereby obtaining the carrier image sequence C = {c 1 ,c 2 ,…,c n}.

[0084] Furthermore, the steganalysis fitness function F of the image set I is s (I) is defined as formula (5):

[0085]

[0086] S105: Encode the secret message and fragment the encoded message;

[0087] Specifically, the user can select the encoding method of the secret message as needed. In this implementation, the joint source-channel encoding method is selected, which mainly includes: first, in order to enhance randomness, the secret message M is encoded. a Encrypt and scramble to get the sequence and Then, after the number system conversion, the message fragments are calculated using data decomposition based on the Vandermonde matrix Improve the integrity and reliability of the extraction. Subsequently, error correction coding, such as RS and BCH coding algorithms, is performed on each message fragment to further improve robustness. Finally, the encoded message fragment is obtained. k m′ and l m′ Indicates the number and length of message fragments.

[0088] S106: solving an optimal payload sequence; the optimal payload sequence refers to a sequence in which the lengths of message segments to be embedded in all carrier images are arranged in the order of carrier images under the condition that the embedding traces of the carrier image sequence are minimized;

[0089] It can be understood that by minimizing the embedding traces as the generation condition of the payload sequence, the concealment of the steganography method can be effectively improved, and the anti-detection and communication reliability of the entire steganography method can be improved.

[0090] Specifically, first, for the obtained carrier image sequence C = {c 1 ,c 2 ,c 3 ,…,c n Each image c in i , the maximum effective load ω can be calculated i .

[0091] Then, based on the maximum payload ω i, considering the changes of information embedding on the high-dimensional features of the image, and measuring the number of groups k in various embedded message fragments i The corresponding carrier image embedding trace E s , as shown in formula (6).

[0092]

[0093] It can be understood that the above formula (6) is essentially consistent with formula (3). The meanings of some parameters can be referred to in formula (3) and will not be repeated here. Represented in the carrier image c i Embed k i The encrypted image after the group message is fragmented. Represented in the carrier image c i Embed k i The embedded traces after the group message is fragmented, l m′ Indicates the length of a single message fragment. Indicates from The features extracted from k i = 0 means no message is embedded in c i In the above example, +∞ indicates that the embedding fails because the message length exceeds the maximum payload.

[0094] Next, the optimal load distribution problem is transformed into the embedding trace minimization problem under load constraints, as shown in formula (7).

[0095]

[0096] Among them, k m′ represents the total number of message fragments, and n represents the total number of carrier images.

[0097] Finally, the solution is obtained by constructing a local optimal sub-model and state recursion. Figure 3 As shown, for the carrier image set C i ={c 1 ,c 2 ,c 3 ,…,c i}, and deduce its effective load k according to the following recursive calculation j The minimum embedding trace at And record the corresponding q value at this time as G s [C i ,k j ]; where 1≤k j ≤k m′ ;

[0098]

[0099] in, Initialize to q represents the optimal result in the current stage corresponding to the carrier image set C i The number of message fragments embedded in ;

[0100] The optimal payload sequence p = {ρ 1 ,ρ 2 ,ρ 3 ,…,ρ n}, ρ i =q×l m′ .

[0101] S107: generating a encrypted image sequence based on a minimum embedding cost algorithm according to the carrier image sequence and the optimal payload sequence;

[0102] Specifically, based on the carrier image sequence C and the payload sequence p, a robust steganography algorithm (such as JCRISBE, MREAS, etc.) is first used to extract the robust carrier elements and the embedding modification amplitude. Then, the embedding cost of each robust carrier element is calculated using the robust-distortion cost function that comprehensively considers the modification amplitude and direction. Finally, the robust carrier elements in each carrier image are integrated and combined with the encoded message fragments. The robust cryptanalyst elements are obtained using the minimum embedding cost code (such as STCs), and the cryptanalyst image sequence S = {s 1 ,s 2 ,s 3 ,…,s n}.

[0103] S108: Publish the secret image sequence using a social network so that a receiver can download and extract the secret message.

[0104] Specifically, the process of extracting the secret message is as follows: (1) Robust steganographic extraction: According to a predetermined robust steganographic algorithm, the coefficients of the embedded message are extracted from each steganographic image of the sequence S, and the robust steganographic elements for secret message recovery are calculated and constructed.

[0105] (2) Secret message extraction: The robust secret element is decoded using the minimum embedding cost code to obtain the message fragment embedded in each secret image. Then, the extracted message fragments are merged and sorted according to the length l. m′ Grouping to get the encoded message fragments

[0106] (3) Secret message reconstruction: First, use the error-correcting code to fragment the extracted message. Decode and get Then, data is reconstructed based on the Vandermonde matrix to restore the message sequence On this basis, reverse scrambling and decryption are performed to obtain the extracted secret message sequence and

[0107] The batch image steganography method applicable to social networks provided by the embodiment of the present invention provides a new idea for describing steganalysis embedding traces, and directly improves the ability to resist steganalysis by utilizing highly separable steganalysis feature selection and difference measurement.

[0108] like Figure 4 As shown, an embodiment of the present invention also provides a batch image steganography system suitable for social networks, including: a preprocessing module, an embedding trace measurement module, a steganographic fitness calculation module, a carrier selection module, a message encoding module, a load distribution module, an embedding module and a communication module.

[0109] The preprocessing module is used to obtain a set of candidate carrier images, perform steganographic processing on the candidate carrier image set using a robust steganographic algorithm, and generate a corresponding set of steganographic images; the embedding trace measurement module is used to measure the embedding trace of the robust steganographic algorithm on the candidate carrier image for any candidate carrier-carrier image pair; the steganalysis fitness calculation module is used to calculate the steganalysis fitness of any candidate carrier image according to its embedding trace; wherein, the smaller the value of the average embedding trace of a candidate carrier image under different payloads, the higher the corresponding steganalysis fitness value; the carrier selection module is used to select an image with a higher steganalysis fitness value from the candidate carrier image set The carrier image sequence is constructed; the message encoding module is used to encode the secret message and segment the encoded message; the load distribution module is used to solve the optimal effective load sequence; the optimal effective load sequence refers to a sequence in which the lengths of the message segments to be embedded in all carrier images are arranged in the order of the carrier images under the condition that the embedding traces of the carrier image sequence are minimized; the embedding module is used to generate a secret image sequence based on the minimum embedding cost algorithm according to the carrier image sequence and the optimal effective load sequence; the communication module is used to publish the secret image sequence through social networks for the receiver to download and extract the secret message.

[0110] It should be noted that the batch image steganography system provided in the embodiment of the present invention is for realizing the above method. The specific functions thereof can be referred to the above method embodiment and will not be described in detail here.

[0111] In order to verify the effectiveness of the solution of the present invention (hereinafter referred to as Multi-Steg), the present invention also provides the following experimental data.

[0112] 1. Experimental setup

[0113] Using 10,000 images from the BOSSbase1.01 image library and digital images taken by a Canon EOS R6 camera, after JPEG compression with a quality factor of 85, we constructed a candidate vector image set I. g and I c Then, the comparison algorithm is constructed with the help of four typical online social network robust information hiding algorithms, including four representative carrier selection algorithms and three latest batch steganographic payload allocation algorithms. Subsequently, the above robust steganographic algorithms are used to embed randomly generated secret messages in combination with multiple batch steganographic models to generate the corresponding secret images.

[0114] In order to test the stealthiness of communication, the PSNR value of the cryptanalysis image, the detection error rate of three typical steganalysis features and the representative deep learning-based steganalysis detection algorithm are used as evaluation indicators. In terms of communication reliability and integrity evaluation, the accuracy of secret message extraction and the number of cryptanalysis images that achieve reliable communication are used as measurement criteria. On this basis, we give the comparison results with representative carrier selection and load distribution algorithms, and use social networks for transmission tests to demonstrate the communication capabilities of the proposed model and method in real situations.

[0115] The specific experimental settings are shown in Table 1.

[0116] Table 1 Experimental settings

[0117] Source Image BOSSbase 1.01database / Canon EOS R6 Image size 512×512 / 4680×3072 Carrier Set <![CDATA[Grayscale Image I g / Color Image I c > Number of carriers 10000+10000＝20000 Number of secrets 20000×10×4×10＝8000000 Secret Message Randomly generated binary sequence Load 0.01,0.03,...,0.09bpnzAC / 0.1,0.2,...,0.5bpnzAC Robust Information Hiding Algorithm JCRISBE(BCH(127,8)) / QTRS / GMAS(RS(31,15)) / RWOSN Vector selection algorithm Rand / Means / Variences / ITC Load distribution strategy UD / DD / ITC Batch Steganography Model Rand-UD / DD-GD / ITC-GD Steganalysis Algorithm DCTR / GFR / J+SRM / SRNet

[0118] (II) Experimental results

[0119] (1) Vector selection test

[0120] From the vector image set I g After generating robust steganalysis images under different loads, we first select 5000 images with the highest steganalysis fitness using the embedding trace function according to the difference between the original steganalysis features and the simplified steganalysis features before and after embedding, and record them as Fitness and Fitness-R respectively. Then, we select 5000 images with the smallest mean and variance of the embedding traces and the largest image texture complexity (ITC). On this basis, we calculate the anti-detection performance of the selected images against the commonly used steganalysis detection features, as shown in Tables 2 and 3. The detection error rate (Rand) of 5000 randomly selected steganalysis images is also given in the table for comparison.

[0121] Table 2 Average detection error rate (0.01, 0.03, ..., 0.09 bpnzAC)

[0122]

[0123]

[0124] Table 3 Average detection error rate (0.1, 0.2, ..., 0.5 bpnzAC)

[0125]

[0126]

[0127] From the results, it can be seen that compared with the most advanced algorithms, Multi-Steg's anti-detection performance for three typical steganalysis features is improved by 10.03%, 10.48% and 11.29% respectively, which shows that the model has stronger confidentiality in realizing covert communication in online social networks.

[0128] (2) Load distribution test

[0129] To compare the performance of payload distribution, we g 5000 carrier images are randomly selected (Rand) and sorted according to embedding distortion (DD), texture complexity (ITC) and steganalysis fitness (Fitness-R). Then, the uniform (UD), greedy (GD) and minimum embedding trace (MES) strategies are used to distribute and embed secret messages under different payloads. Subsequently, the anti-detection performance of the generated secret images is tested by some typical high-dimensional steganalysis features and deep learning-based steganalysis algorithms. The results are shown in Tables 4, 5 and Figure 5 shown.

[0130] Table 4 Average detection error rate (0.01, 0.03, ..., 0.09 bpnzAC)

[0131]

[0132]

[0133] Table 5 Average detection error rate (0.1, 0.2, ..., 0.5 bpnzAC)

[0134]

[0135]

[0136] Experimental results show that the detection error rate of the steganographic images generated by the representative steganalysis detection features and SRNet, Multi-Steg model is always higher than 49.24%, which is significantly improved compared with batch steganographic models such as Rand-UD, DD-GD and ITC-GD

[13] .

[0137] (1) Social network transmission test

[0138] Considering the actual application scenarios, from I c 200 images are randomly selected as carriers. Then, RWOSN is used to embed the secret message by combining uniform embedding and Multi-Steg. After propagation through Facebook and Weibo, the average PSNR between the carrier image and the received image, the message extraction error rate, and the number of images from which the message can be correctly extracted from the received image are calculated, as shown in Tables 6 and 7.

[0139] Table 6 PSNR mean, extraction error rate and number of correct images (0.01, 0.03, ..., 0.09 bpnzAC)

[0140]

[0141] Table 7 PSNR mean, extraction error rate and number of correct images (0.1, 0.2, ..., 0.5 bpnzAC)

[0142]

[0143] Experimental results show that compared with RWOSN, the PSNR value of the Multi-Steg model generated secret images increased by 4.68-11.04 dB, the message extraction error rate decreased by 2.12-4.26 times, and the communication success rate increased by 32.5%-61%. In summary, using Multi-Steg to transmit secret messages on Facebook and Weibo can significantly improve image quality and message integrity, providing an effective solution for reliable secret communication in public channels.

[0144] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor (processor) 601, a communication interface (Communications Interface) 602, a memory (memory) 603 and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604. The processor 601 can call the logic instructions in the memory 603 to execute the batch image steganography method, which includes: obtaining a candidate carrier image set, performing steganographic processing on the candidate carrier image set using a robust steganography algorithm, and generating a corresponding secret image set; for any candidate carrier-secret image pair, measuring the embedding trace of the robust steganography algorithm on the candidate carrier image; for any candidate carrier image, calculating its steganographic fitness according to its embedding trace; wherein, the smaller the value of the average embedding trace of a candidate carrier image under different payloads, the higher the corresponding steganographic fitness value; selecting images with higher steganographic fitness values ​​from the candidate carrier image set to construct a carrier image sequence; encoding the secret message and fragmenting the encoded message; solving the optimal payload sequence; the optimal payload sequence refers to a sequence in which the lengths of the message fragments to be embedded in all carrier images are arranged in the order of the carrier images under the condition that the embedding trace of the carrier image sequence is minimized; generating a secret image sequence based on the carrier image sequence and the optimal payload sequence based on the minimum embedding cost algorithm; and publishing the secret image sequence using a social network for the receiver to download and extract the secret message.

[0145] In addition, when the logic instructions in the above-mentioned memory 603 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as 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.

[0146] The embodiment of the present invention further provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments, for example, including: obtaining a set of candidate carrier images, performing steganographic processing on the candidate carrier image set using a robust steganographic algorithm, and generating a corresponding set of secret images; for any candidate carrier-secret image pair, measuring the embedding traces of the robust steganographic algorithm on the candidate carrier image; for any candidate carrier image, calculating its steganographic fitness based on its embedding traces; wherein a candidate carrier image has different embedding traces. The smaller the value of the average embedding trace under the effective load, the higher the corresponding steganographic fitness value; select images with higher steganographic fitness values ​​from the candidate carrier image set to construct a carrier image sequence; encode the secret message and segment the encoded message; solve the optimal effective load sequence; the optimal effective load sequence refers to a sequence in which the lengths of the message segments to be embedded in all carrier images are arranged in the order of carrier images under the condition that the embedding trace of the carrier image sequence is minimized; generate a secret image sequence based on the minimum embedding cost algorithm according to the carrier image sequence and the optimal effective load sequence; use social networks to publish the secret image sequence for the receiver to download and extract the secret message.

[0147] The embodiment of the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method provided by each of the above method embodiments is implemented, for example, including: obtaining a set of candidate carrier images, performing steganographic processing on the candidate carrier image set using a robust steganographic algorithm, and generating a corresponding set of secret image sets; for any candidate carrier-secret image pair, measuring the embedding trace of the candidate carrier image by the robust steganographic algorithm; for any candidate carrier image, calculating its steganographic fitness according to its embedding trace; wherein, the smaller the value of the average embedding trace of a candidate carrier image under different payloads, the corresponding The higher the value of the steganographic fitness is; select images with higher steganographic fitness values ​​from the candidate carrier image set to construct a carrier image sequence; encode the secret message and segment the encoded message; solve the optimal payload sequence; the optimal payload sequence refers to a sequence in which the lengths of the message segments to be embedded in all carrier images are arranged in the order of the carrier images under the condition that the embedding traces of the carrier image sequence are minimized; generate a secret image sequence based on the minimum embedding cost algorithm according to the carrier image sequence and the optimal payload sequence; use social networks to publish the secret image sequence for the recipient to download and extract the secret message.

[0148] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A batch image steganography method suitable for social networks, characterized in that: include: Step 1: Obtain a candidate carrier image set, use a robust steganography algorithm to perform steganographic processing on the candidate carrier image set, and generate a corresponding steganographic image set; Step 2: for any candidate carrier-stamped image pair, measure the embedding trace of the robust steganography algorithm on the candidate carrier image; Step 3: For any candidate carrier image, calculate its steganalytic fitness according to its embedding trace; the smaller the average embedding trace value of a candidate carrier image under different payloads, the higher the corresponding steganalytic fitness value; Step 4: Select images with higher steganographic fitness values ​​from the candidate carrier image set to construct a carrier image sequence; Step 5: Encode the secret message and fragment the encoded message; Step 6: solving the optimal payload sequence; the optimal payload sequence refers to a sequence in which the lengths of message segments to be embedded in all carrier images are arranged in the order of carrier images under the condition that the embedding traces of the carrier image sequence are minimized; Step 7: Generate a encrypted image sequence based on a minimization embedding cost algorithm according to the carrier image sequence and the optimal payload sequence; Step 8: Utilize social networks to publish the secret image sequence for the recipient to download and extract the secret message.

2. According to claim 1, a batch image steganography method suitable for social networks is characterized in that: Step 2 specifically includes: For any candidate carrier-secret image pair, extract key features from the candidate carrier image and the corresponding secret image respectively, wherein the key features refer to features that can distinguish the candidate carrier from the corresponding secret image; According to the key features extracted from the candidate carrier image and the corresponding steganographic image, the embedding trace of the candidate carrier image by the robust steganographic algorithm is measured as follows: in, and From the candidate carrier images and the corresponding encrypted image The key features extracted from c (·) and τ s (·)respectively and The average value of all characteristic components in c (·) and υ s (·) respectively and The standard deviation of all feature components in .

3. A batch image steganography method suitable for social networks according to claim 2, characterized in that: For any candidate carrier-secret image pair, extract the key features of the candidate carrier image and the corresponding secret image, including: The characteristic component f is calculated as follows *,j Divisibility of A s (f *,j ); Among them, μ c (·) and μ s (·) are the average values ​​of the j-th dimension feature components of all candidate carriers and the corresponding encrypted images, δ c (·) and δ s (·) are the standard deviations of the j-th dimension feature components of all candidate carriers and the corresponding encrypted images; For each feature component, if its separability value is greater than the set separability threshold, the feature component is added to the key feature candidate set; based on the current key feature candidate set, the candidate carrier image and the secret carrier image are classified, if the classification accuracy is greater than the set accuracy threshold, the feature component is added to the key feature set, otherwise the feature component is deleted from the key feature candidate set.

4. A batch image steganography method suitable for social networks according to claim 1, characterized in that: In step 3, calculate any candidate carrier image according to the following formula The steganalytic fitness in, is the candidate carrier image The embedding trace, ρ is α1~α h Different loads in the range, λ i is the candidate carrier image Average embedding traces under different effective loads, h represents the number of values ​​of different loads.

5. The batch image steganography method suitable for social networks according to claim 1, characterized in that: Step 6 specifically includes: Calculate each carrier image c i The maximum payload is calculated when the robust steganography algorithm is used on the carrier image c i Embed k i The embedded traces of group messages after fragmentation; Solving the optimal payload sequence is transformed into the following optimization problem; in, Represented in the carrier image c i Embed k i The encrypted image after the group message is fragmented. Represented in the carrier image c i Embed k i The embedding traces after group message fragmentation, k m′ Indicates the total number of message fragments, l m′ Indicates the length of a single message segment, and n indicates the total number of carrier images; A multi-stage decision optimization is adopted to solve the optimization problem and obtain the optimal payload sequence.

6. A batch image steganography method suitable for social networks according to claim 5, characterized in that: The optimization problem is solved by multi-stage decision optimization, which specifically includes: For the carrier image set C i ={c1,c2,c3,…,c i }, and deduce its effective load k according to the following recursive calculation j The minimum embedding trace at And record the corresponding q value at this time as G s [C i ,k j ]; where 1≤k j ≤k m′ ; in, Initialize to q represents the optimal result in the current stage corresponding to the carrier image set C i The number of message fragments embedded in ; The number of message fragments embedded in the carrier image corresponding to the optimal result in each stage is traced back to obtain the optimal payload sequence p = {ρ1, ρ2, ρ3, …, ρ n }, ρ i =q×l m′ .

7. A batch image steganography system suitable for social networks, characterized in that: include: The preprocessing module is used to obtain a candidate carrier image set, perform steganographic processing on the candidate carrier image set using a robust steganographic algorithm, and generate a corresponding steganographic image set; An embedding trace measurement module, used for measuring the embedding trace of the robust steganography algorithm on any candidate carrier-stamped image pair; The stegographic fitness calculation module is used to calculate the stegographic fitness of any candidate carrier image according to its embedding trace; wherein the smaller the average embedding trace value of a candidate carrier image under different payloads is, the higher the corresponding stegographic fitness value is; A carrier selection module is used to select images with higher steganographic fitness values ​​from a candidate carrier image set to construct a carrier image sequence; A message encoding module, used to encode the secret message and fragment the encoded message; A load distribution module is used to solve the optimal effective load sequence; the optimal effective load sequence refers to a sequence in which the lengths of message segments to be embedded in all carrier images are arranged in the order of carrier images under the condition that the embedding traces of the carrier image sequence are minimized; An embedding module, used for generating a sequence of encrypted images based on a minimum embedding cost algorithm according to the carrier image sequence and the optimal payload sequence; The communication module is used to publish the secret image sequence using a social network so that a receiver can download and extract the secret message.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images