An adaptive immune steganographic image construction method

Abstract

This invention discloses a method for constructing immune steganographic images based on an artificial immune system, suitable for digital image information hiding. By constructing an information hiding model based on the artificial immune system, the method automatically optimizes and modifies the secret image with the goal of reducing the distance between the residuals of the carrier image and the secret image to generate an immune steganographic image, thereby effectively enhancing the security of the steganographic algorithm. The method is universal and applicable to any steganographic method that minimizes distortion. It ensures that modifying the stegoimage does not cause secret message extraction failures and does not introduce additional detectable artifacts, further improving the security performance of existing steganographic algorithms and possessing practical value.

Description

Technical Field

[0001] The present invention relates to the field of multimedia security, and in particular to an immune steganographic image construction algorithm based on an artificial immune system, which improves the security of resisting steganalysis while ensuring accurate extraction of hidden data. Background Art

[0002] Steganography is the art and science of covert communication, aiming to conceal secret information within seemingly natural digital media without arousing the suspicion of steganalysts. It is typically achieved by modifying the carrier, such as the pixel values ​​of spatial domain images and the DCT coefficients of JPEG images. The fundamental task of steganalysis, as the adversary in the game, is to uncover the presence of secret information within the stegomedia by searching for anomalous artifacts caused by data embedding. Among all steganographic techniques, image steganography plays a significant role in covert communication and has attracted increasing attention in recent years. Currently, state-of-the-art adaptive image steganography schemes are designed using a distortion minimization framework to minimize statistical detectability by steganalysis.

[0003] Current adaptive steganography algorithms, such as HUGO (Using high-dimensional image models to perform highly undetectable steganography, International Workshop on Information Hiding, 2010, pp. 161-177), S-UNIWARD (Digital image steganography using universal distortion, Proceedings of the first ACM Workshop on Information Hiding and Multimedia Security, 2013, pp. 59-68), HILL (A new costfunction for spatial image steganography, 2014 IEEE International Conference on Image Processing (ICIP), 2014, pp.4206-4210), MIPOD (Content-adaptivesteganography by minimizing statistical detectability, IEEE Transactions onInformation Forensics and Security, 2016, vol.11, no.2, pp.221-234) and UT-GAN (Anembedding cost learning framework using GAN,IEEE Transactions on Information Forensics and Security, 2020, vol. 15, pp. 839-851) all design distortion cost functions and calculate the distortion cost of each pixel to minimize the overall distortion of the steganographic image after data embedding. However, none of them take into account that after the secret message is embedded, immunizing the carrier image can improve the security of the steganography, and with the emergence of more and more powerful steganalyzers, the security of these steganographic schemes that only rely on the design of distortion functions is getting lower and lower. Since the most advanced steganalyzers are mainly based on the residual of the image for analysis, it is reasonable to reduce or eliminate the distance between the residual of the carrier image and the residual of the steganalyzer by modifying the steganalyzer image to resist steganalysis. However, simply modifying the steganalyzer image may cause the failure of secret message extraction and introduce additional detectable artifacts.The artificial immune system is a self-learning and adaptive intelligent computing method that obtains the optimal solution by constructing immune evolution. Based on the artificial immune system, this paper proposes an adaptive immune steganographic image construction method to further enhance the security of existing steganography. Summary of the Invention

[0004] The purpose of the present invention is to solve the above problems and provide an immune steganographic image construction algorithm based on an artificial immune system. The algorithm uses the artificial immune system to automatically modify the steganographic image after embedding the secret message to minimize the Euclidean distance between the residual co-occurrence matrix features of the carrier image and the modified steganographic image to generate the immune steganographic image, thereby effectively improving the security performance of the existing steganographic algorithm.

[0005] The technical solutions for achieving the purpose of the present invention are as follows:

[0006] A method for constructing an immune steganographic image based on an artificial immune system is used for digital image information hiding. The method performs immune processing on a secret image obtained by an existing steganographic method based on the artificial immune system to form an immune steganographic image, thereby effectively enhancing the security of the steganographic algorithm. The method comprises the following steps:

[0007] Step 1: Generate a secret image using a basic steganographic algorithm;

[0008] Step 2: Perform immune processing on the steganographic image generated in step 1 to form an immune steganographic image.

[0009] The specific contents of step 1 above are:

[0010] Step 1-1, use the existing steganography method to calculate the carrier X i The cost matrix C of the elements in i ;

[0011] Step 1-2, use STCs according to the distortion cost matrix C i Embed the secret message into carrier X i , forming a cipher image Y containing a secret message i .

[0012] The specific content of step 2 above is:

[0013] Step 2-1, encrypted image Y i Encoded as the initial antibody Ab i , forming the initial antibody population Pop(0);

[0014] Step 2-2, calculate the affinity of each antibody Fit (Ab i ), to assess whether it is immune selected;

[0015] In step 2-3, antibodies with a ratio of d1 are selected for cloning based on affinity, and the selected antibodies are cloned according to the set cloning rate Ncl to form a clone pool.

[0016] In steps 2-4, all cloned antibodies are mutated using a mutation operator to generate new antibodies.

[0017] Steps 2-5: Replace the parent antibody with a mutant antibody with high affinity to form an antibody population Pop new (t);

[0018] Step 2-6, generating d2 new antibodies in the same manner as forming initial antibodies to replace the same number of antibodies in the antibody population, thereby generating a new antibody population Pop(t+1);

[0019] Step 2-7, repeat steps 2-2 to 2-7 until the set end condition is met, and the best antibody Ab with the highest affinity is output best , that is, immune steganographic image Z best .

[0020] The specific contents of step 2-1 above are:

[0021] Step 2-1A, use the following encoding method to encode the encrypted image into antibody Ab i :

[0022]

[0023] Among them, Ab i Y is encoded as an antibody i , ab q (ab q =y i,j ) represents the qth (q=(i-1)×w+j) element, where h and w are the height and width of the image.

[0024] Step 2-1B, given an initial population size of n, replicate n Abs i , and then use the mutation operator to mutate the antibody to obtain the initial antibody population Pop(0):

[0025] Pop(0)={Ab i (0)|i∈[1,n]}

[0026] Among them, the designed mutation operator is as follows:

[0027] Ab=ab i +4×k,i∈rand([1,h×w],1) and Ab=ab i +4×k,i∈rand([1,h×w],t)

[0028] Among them, rand([min,max],t) randomly selects t integers in [min,max], k∈{-1,0,+1}.

[0029] The specific contents of step 2-2 above are:

[0030] In step 2-2A, the residual between the carrier X and the modified stego-image Z (i.e., the antibody) is calculated using the following residual calculation method:

[0031] and

[0032] in,

[0033] In step 2-2B, the following truncation operation is used to fix the changes of the residual images Res(X) and Res(Z) to [-T v ,T v ]:

[0034]

[0035] and

[0036]

[0037] Among them, trunc(*) is a truncation operation;

[0038] In step 2-2C, the co-occurrence matrix of the adjacent pixels of the residual matrix is ​​calculated along the horizontal, vertical, diagonal, and anti-diagonal directions using the following calculation method:

[0039]

[0040]

[0041]

[0042]

[0043] Among them, r i,j is the pixel at position (i, j) of the residual image, k,l∈[-T v ,T v ],

[0044] In step 2-2D, the statistical features of the residual image are calculated as follows:

[0045] and

[0046] In step 2-2E, the affinity of each antibody is calculated using the following affinity function:

[0047]

[0048] Where r is F total Dimensions, and are the statistical characteristics of the carrier X and the modified stego-image Z (i.e., the antibody).

[0049] The specific contents of steps 2-3 above are:

[0050] Step 2-3A, sort all antibodies in non-increasing order of affinity, and select d1×n antibodies with higher affinity;

[0051] In step 2-3B, the selected antibodies are cloned as follows:

[0052] cAb=[clone(Ab1),clone(Ab2),...,clone(Ab d1×n )]

[0053] Among them, clone(Ab j )=I j ×Ab j , j=1,2,....,d1×n,I j is a row vector of Ncl dimension, where Ncl is the clone size of the antibody.

[0054] The specific contents of the above steps 2-4 are: mutating each cloned antibody using the following mutation method:

[0055] Ab=ab i +4×k,i∈rand([1,h×w],1)

[0056] The specific contents of the above steps 2-5 are: using affinity function to calculate the Ab of each mutant antibody i The affinity of the antibody is then replaced by the mutant antibody with higher affinity to form a new antibody population Pop new (t).

[0057] The specific contents of steps 2-6 above are:

[0058] In step 2-6A, d2×n new antibodies are generated as follows:

[0059] Ab=ab i +4×k,i∈rand([1,h×w],1) and Ab=ab i +4×k,i∈rand([1,h×w],t)

[0060] Step 2-6B, merge these d2×n new antibodies into Pop new (t), forming the next generation antibody population Pop(t+1) and entering the next iteration.

[0061] The specific contents of the above steps 2-7 are: set the maximum number of iterations T max = 200, repeat the above steps 2-2 to 2-6, when t = T max , terminate the operation and output the optimal immune steganographic image Z best .

[0062] The core concept of the proposed immune stego-image construction algorithm based on an artificial immune system is to use the artificial immune system to automatically modify the steganographic image embedded with the secret message to minimize the Euclidean distance between the residual co-occurrence matrix features of the carrier image and the modified steganographic image, thereby generating the immune stego-image. Compared with existing image-adaptive steganography algorithms, this algorithm offers advantages in achieving higher security than existing steganography algorithms and is universally applicable. It can be combined with steganographic images generated by any steganographic algorithm, consistently improving security against various steganalyzers. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 An embodiment of the present invention provides an immune steganographic image construction framework based on an artificial immune system.

[0064] Figure 2 This is an example diagram of the initial population according to an embodiment of the present invention.

[0065] FIG3 is a diagram illustrating examples of antibody mutations according to an embodiment of the present invention.

[0066] FIG3( a ) is an example diagram of a single-point variation according to an embodiment of the present invention.

[0067] FIG3( b ) is a diagram illustrating a multi-point variation example according to an embodiment of the present invention.

[0068] Figure 4 This is a modified diagram of an embodiment of the present invention.

[0069] Figure 5 This is an example diagram of the original carrier and the corresponding immune steganographic image according to an embodiment of the present invention.

[0070] Figure 6 The experimental results of the embodiment of the present invention and the existing method under SRM detection are compared.

[0071] Figure 7 The figure compares the experimental results of the embodiment of the present invention and the existing method under maxSRMd2 detection.

[0072] Figure 8 The experimental results of the embodiment of the present invention and the existing method under Zhu-Net detection are compared. DETAILED DESCRIPTION

[0073] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0074] like Figure 1 As shown, the present invention provides a method for constructing an immune steganographic image based on an artificial immune system, comprising the following steps:

[0075] Step 1-1, use the existing steganography method to calculate the carrier X i The cost matrix C of the elements in i ;

[0076] Step 1-2, use STCs according to the distortion cost matrix C i Embed the secret message into carrier X i , forming a cipher image Y containing a secret message i .

[0077] Step 2-1A: In order to determine the encrypted image as an antibody, it is necessary to encode the encrypted image. The following encoding method is used to encode the encrypted image into an antibody Ab. i :

[0078]

[0079] Among them, Ab i Y is encoded as an antibody i , ab q (ab q =y i,j ) represents the qth (q=(i-1)×w+j) element, where h and w are the height and width of the image.

[0080] Population initialization marks the initial implementation of the artificial immune system. The initial population affects the security of the immune image and the convergence speed of the artificial immune algorithm. Randomly generating the initial population may result in an inability to correctly extract the message from the resulting immune steganographic image. Therefore, the population initialization of this method is described as follows.

[0081] Step 2-1B, given an initial population size of n, replicate n Abs i , and then use the mutation operator to mutate the antibody to obtain the initial antibody population Pop(0), such as Figure 2 As shown:

[0082] Pop(0)={Ab i (0)|i∈[1,n]}

[0083] Among them, the designed mutation operator is as follows:

[0084] Ab=ab i +4×k,i∈rand([1,h×w],1) and Ab=ab i +4×k,i∈rand([1,h×w],t)

[0085] Among them, rand([min,max],t) randomly selects t integers in [min,max], k∈{-1,0,+1}.

[0086] As shown in Figure 3, using the designed single-point mutation operator and multi-point mutation operator, it can be ensured that the secret message can be accurately extracted in individuals using STCs.

[0087] Antibody affinity is designed to assess the strengths or weaknesses of an antibody. The design of the affinity function should be closely related to the specific target problem, which is defined based on the characteristics of different optimization objectives. Furthermore, this method aims to generate more secure stego images, and the antibody affinity should also be related to the undetectability of the modified stego images. The affinity function used in this method is described below.

[0088] In step 2-2A, the residual between the carrier X and the modified stego-image Z (i.e., the antibody) is calculated using the following residual calculation method:

[0089] and

[0090] in,

[0091] For obtaining image residuals, a filter group composed of multiple filters may produce more complex image residuals, but too many convolution operations will inevitably increase the time required for affinity calculation. A single KV filter can produce relatively rich image residuals while avoiding wasting computing time.

[0092] In step 2-2B, the following truncation operation is used to fix the changes of the residual images Res(X) and Res(Z) to [-T v ,T v ]:

[0093]

[0094] and

[0095]

[0096] Among them, trunc(*) is a truncation operation;

[0097] In step 2-2C, the co-occurrence matrix of the adjacent pixels of the residual matrix is ​​calculated along the horizontal, vertical, diagonal, and anti-diagonal directions using the following calculation method:

[0098]

[0099]

[0100]

[0101]

[0102] Among them, r i,j is the pixel at position (i, j) of the residual image, k,l∈[-T v ,T v ],

[0103] In step 2-2D, the statistical features of the residual image are calculated as follows:

[0104] and

[0105] In step 2-2E, the affinity of each antibody is calculated using the following affinity function:

[0106]

[0107] Where r is F total Dimensions, and are the statistical characteristics of the carrier X and the modified stego-image Z (i.e., the antibody).

[0108] Step 2-3A, sort all antibodies in non-increasing order of affinity, and select d1×n antibodies with higher affinity;

[0109] In step 2-3B, the selected antibodies are cloned as follows:

[0110]

[0111] Among them, clone(Ab j )=I j ×Ab j , j=1,2,....,d1×n,I j is a row vector of Ncl dimension, where Ncl is the clone size of the antibody.

[0112] 2-4: Mutate each cloned antibody using the following mutation method:

[0113] Ab=ab i+4×k,i∈rand([1,h×w],1)

[0114] Step 2-5: Calculate the Ab of each mutant antibody using affinity function i The affinity of the antibody is then replaced by the mutant antibody with higher affinity to form a new antibody population Pop new (t).

[0115] In step 2-6A, d2×n new antibodies are generated as follows:

[0116] Ab=ab i +4×k,i∈rand([1,h×w],1) and Ab=ab i +4×k,i∈rand([1,h×w],t)

[0117] Step 2-6B, merge these d2×n new antibodies into Pop new (t), forming the next generation antibody population Pop(t+1) and entering the next iteration.

[0118] Step 2-7: Set the maximum number of iterations T max , repeat the above steps 2-2 to 2-6, when t=T max , terminate the operation and output the optimal immune steganographic image Z best .

[0119] In this example, the population size is set to 30, the maximum number of iterations is set to 100, the selection ratio d1 is set to 0.1, the number of antibody clones Ncl is set to 40, and the cutoff value T is set to 0. v Set to 3, and the update ratio d2 to 0.9.

[0120] Figure 4 Figure 1 shows the modified pixels for images 1013.pgm, 5013.pgm, and 8888.pgm from the BOSSbase dataset at different iteration times in this example. As the number of iterations increases, the adaptive nature of this method becomes increasingly apparent. After 100 iterations, the immune-treated points for images 1013.pgm, 5013.pgm, and 8888.pgm are highly concentrated in the complex texture areas of the stego-image, and the number of immune-treated points increases significantly. This shows that complex texture areas not only have a higher probability of embedding but also a higher probability of immune treatment. Immunity-based stego-immune treatment primarily focuses on complex texture areas.

[0121] Figure 5Figure 2 shows an example of the original vector and the corresponding immune steganographic image from this example. c_xxxx and s_xxxx represent the original vector and immune vector, respectively, and images 1013, 4568, 5013, and 8888 are selected from the BOSSbase dataset. As can be seen, there is no significant difference between the steganographic image and the immune steganographic image, demonstrating the excellent invisibility of this method.

[0122] This embodiment uses three steganalyzers to verify the security performance of this method, namely, the steganalyzers SRM and maxSRMd2 based on manual features, and the steganalyzer Zhu-Net based on neural network. This embodiment uses the average detection error rate P of 10 detection results. E As an evaluation indicator of steganographic security.

[0123] Figure 6 is the average detection error rate P of the present invention and the existing method under SRM E , Figure 7 is the average detection error rate of the present invention and the existing method under maxSRMd2, Figure 8 is the average detection error rate of the present invention and the existing method under Zhu-Net. Figure 6 、 Figure 7 and Figure 8 The results show that the detection error rates of the steganographic images generated by the heuristic steganographic algorithms S-UNIWARD, WOW and HILL, the model-based steganographic algorithm MiPOD and the corresponding immune steganographic images generated by the present invention are resistant to various types of steganography under different payloads. Figure 6 、 Figure 7 and Figure 8 It can be seen that compared with existing steganography schemes, the present invention can achieve significantly higher security against the most advanced hand-crafted feature-based steganalyzers SRM and maxSRMd2 and the deep learning-based steganalyzer Zhu-Net.

Claims

1. An adaptive immune steganographic image construction method for digital image information hiding. Based on an artificial immune system, the secret image obtained by the existing steganographic method is immunized to form an immune steganographic image, thereby effectively enhancing the security of the steganographic algorithm. The method comprises the following main steps: Step 1: Generate the encrypted image Y using the basic steganography algorithm i ; Step 2: Perform immune processing on the steganographic image generated in step 1 to form an immune steganographic image Z best , the specific implementation is as follows: Step 2-1, encrypted image Y i Encoded as the initial antibody Ab i , forming the initial antibody population Pop(0); Step 2-2, calculate the affinity of each antibody Fit (Ab i ), to assess whether it is immune selected; Step 2-3: select antibodies with a ratio of d1 for cloning based on affinity, and clone the selected antibodies according to the set cloning rate Ncl to form a clone pool; Steps 2-4: mutate all cloned antibodies using a mutation operator to generate new antibodies; Steps 2-5: Replace the parent antibody with a mutant antibody with high affinity to form an antibody population Pop new (t); Step 2-6, generating d2 new antibodies in the same manner as forming initial antibodies to replace the same number of antibodies in the antibody population, thereby generating a new antibody population Pop(t+1); Step 2-7, repeat steps 2-2 to 2-7 until the set end condition is met, and the best antibody Ab with the highest affinity is output best , as the immune steganographic image Z best .

2. The adaptive immune steganographic image construction method according to claim 1, wherein: The specific implementation of step 2.1 is as follows: The following encoding method is used to encode the encrypted image into antibody Ab i : Among them, Ab i Y is encoded as an antibody i , ab q (ab q =y i,j ) represents the qth (q = (i-1) × w + j) element, where h and w are the height and width of the image; Given an initial population size of n, replicate n Abs i , and then use the mutation operator to mutate the antibody to obtain the initial antibody population Pop(0): Pop(0)={Ab i (0)|i∈[1,n]} Among them, the designed mutation operator is as follows: Ab = ab i + 4×k, i ∈ rand([1, h×w], 1) and Ab = ab i + 4×k, i ∈ rand([1, h×w], t) Among them, rand([min,max],t) randomly selects t integers in [min,max], k∈{-1,0,+1}.

3. The adaptive immune steganographic image construction method according to claim 1, wherein: The specific implementation of step 2.2 is as follows: The residual of the carrier X and the modified stego image Z, i.e. the antibody, is calculated using the following residual calculation method: and in, The following truncation operation is used to fix the changes of the residual images Res(X) and Res(Z) to [-T v ,T v ]: and Among them, trunc(*) is a truncation operation; The co-occurrence matrix of adjacent pixels of the residual matrix is ​​calculated along the horizontal, vertical, diagonal, and anti-diagonal directions using the following calculation method: Among them, r i,j is the pixel at position (i, j) of the residual image, k,l∈[-T v ,T v ], The statistical features of the residual image are calculated as follows: and The affinity of each antibody was calculated using the following affinity function: Where r is F total Dimensions, and are the statistical characteristics of the carrier X and the modified stego-image Z.

4. The adaptive immune steganographic image construction method according to claim 1, wherein: The mutation method used in step 2.5 is single-point mutation: Ab=ab i +4×k,i∈rand([1,h×w],1).

Images