Inter-block recombination JPEG steganography method based on channel optimization

Abstract

The invention discloses an inter-block recombination JPEG steganography method based on channel optimization, and relates to the field of image steganography, and the steganography method comprises the steps: S1, carrying out the channel optimization processing of a carrier image, and obtaining a sample image, so as to adapt to a compressed environment after uploading; s2, embedding the ECC (Error Correction Code) encryption information into the sample by using the STC (Syndrome-Trellis Codes), so as to obtain an embedded image; s3, the embedded image is modified and spliced to generate a target image, the inter-block recombination JPEG steganography method based on channel optimization has good robustness while the satisfactory safety performance is kept, and the loss of JPEG type pictures on online social software can be effectively avoided.

Description

Technical Field

[0001] The invention relates to the field of image steganography, in particular to an inter-block reorganization JPEG steganography method based on channel optimization. Background Art

[0002] The background technology of the channel-optimized inter-block reorganization JPEG steganography method involves the fields of information security, multimedia content security, and steganography. In network transmission, in order to reduce storage costs and improve transmission efficiency, digital images are usually transmitted after lossy compression. This processing method makes traditional steganography technology face challenges in extracting secret information, because compression may cause the embedded information to be unable to be completely correctly extracted, thereby affecting the effectiveness of covert communication.

[0003] As the most common form of multimedia in network transmission, JPEG steganography has a wide range of applications. In order to improve the security of JPEG image steganography, researchers mainly focus on reducing embedding distortion and maintaining statistical characteristics. For example, by reasonably selecting embedding coefficients and optimizing the coefficient modification direction using genetic algorithms, embedding distortion can be effectively reduced while maintaining the histogram characteristics of the image. In addition, by analyzing the security of the YASS (Yet Another Secure Steganography) steganography algorithm, a method of randomizing the position of the embedded block is proposed to overcome the overlap problem between the embedded block and the JPEG block, thereby better maintaining the statistical characteristics of the image.

[0004] Steganalysis techniques based on JPEG images are also developing. For example, by extracting the statistical features of DCT (discrete cosine transform) blocks and combining them with the joint probability density matrix, researchers are able to more effectively detect the presence of hidden information. This analysis method not only improves the detection rate of typical JPEG steganography, but also shows good detection results at low embedding rates.

[0005] The improvement of JPEG image steganographic security, the development of steganalysis technology, and the research on robust steganography methods combining lossless carriers and robust costs not only improve the compression resistance and anti-detection performance of steganography technology, but also provide a solid foundation for the security of future multimedia content.

[0006] Therefore, there is an urgent need for an inter-block reorganization JPEG steganography method based on channel optimization to solve the above problems. Summary of the invention

[0007] To solve the above problems, the present application provides a block-to-block reorganization JPEG steganography method based on channel optimization, aiming to improve robustness while ensuring security.

[0008] In the present application, a block-to-block reorganization JPEG steganography method based on channel optimization is provided, comprising the following steps:

[0009] S1, performing channel optimization processing on the carrier image to obtain a sample image to adapt to the compression environment after uploading;

[0010] S2, using STC (Syndrome-Trellis Codes) to embed ECC (Error Correction Code) encryption information in the sample to obtain an embedded image;

[0011] S3, modify the embedded image and splice the generated target image.

[0012] Preferably, the specific content of performing channel optimization processing on the carrier image in S1 is:

[0013] applying an additive distortion cost function to the carrier image and identifying a number of modifiable portions of the carrier image;

[0014] The visual impact caused by the modification in the modifiable part is quantified by an additive distortion cost function;

[0015] According to the result of the quantitative calculation, the initial distortion cost of the modifiable part of the carrier image is obtained;

[0016] The sample images are generated by mapping the calculated initial distortion cost onto the carrier image through color encoding.

[0017] Preferably, the expression of STC (Syndrome-Trellis Codes) in S2 is as follows:

[0018]

[0019] Ext = H·P(Y);

[0020] Where Emb(·) and Ext(·) represent the embedding algorithm and the extraction algorithm, respectively. , m is the embedded message, P(Y) represents the image modifiable component, H is the parity check matrix, and C(m) = {z|Hz = m} represents the complement corresponding to m.

[0021] Preferably, the specific steps of modifying the embedded image to generate the target image in S3 are:

[0022] The embedded image is divided into several 3×3 non-overlapping pixel blocks to generate the initial modification pattern;

[0023] Adjust the embedding direction within non-overlapping pixel blocks, traverse the non-zero elements, and generate 2 N embedding candidate patterns;

[0024] Calculate the embedding impact of each candidate pattern and sort them from small to large, and select the candidate pattern with the highest order as the embedding modification pattern;

[0025] The target image is generated by splicing non-overlapping pixel blocks according to the embedded modification pattern.

[0026] Preferably, the distortion cost is continuously adjusted when adjusting the embedding direction within the non-overlapping blocks and an asymmetric distortion cost adjustment strategy and an asymmetric distortion cost adjustment strategy are applied to the adjustment changes of the non-overlapping pixel blocks;

[0027] The asymmetric distortion cost adjustment strategy includes scaling the initial distortion cost according to the pixel modification direction.

[0028] Preferably, the expression of the asymmetric distortion cost adjustment strategy is:

[0029]

[0030] Among them, the initial distortion cost of the “+1” and “-1” changes of each pixel is symmetrical, that is, ρ + =ρ - =ρ, d0 is the modified pixel x i,j of and, is the optimal change direction pixel x after the change i,j and the asymmetric distortion cost after the changes of “+1” and “-1”, α is the scaling factor (α>1), and the initial distortion cost ρ 0 Modified by "0" for pixels or DCT coefficients, defaults to 0.

[0031] In summary, the block-by-block reorganization JPEG steganography method based on channel optimization of the present invention, compared with the traditional steganography, firstly channels the cover image, then embeds the ECC encryption information through STC, then crops the image containing the information into a series of non-overlapping blocks, and optimizes the cropped blocks again, and selects the optimal blocks for splicing. Experiments show that the method has good robustness while maintaining satisfactory security performance.

[0032] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flow chart of a block-to-block reorganization JPEG steganography method based on channel optimization of the present invention;

[0034] Figure 2 The security comparison of various methods with initial distortion function J-UNIWARD under QF=70 of the inter-block reorganization JPEG steganography method based on channel optimization of the present invention is presented;

[0035] Figure 3 The security comparison of various methods with initial distortion function J-UNIWARD under QF=80 of the block-to-block reorganization JPEG steganography method based on channel optimization of the present invention is presented;

[0036] Figure 4 The security comparison of various methods with initial distortion function UERD under the condition of QF=70 of the inter-block reorganization JPEG steganography method based on channel optimization of the present invention is provided;

[0037] Figure 5 The security comparison of various methods with initial distortion function UERD under the condition of QF=80 of the inter-block reorganization JPEG steganography method based on channel optimization of the present invention is shown. DETAILED DESCRIPTION

[0038] The technical solution of the present invention is further described below by means of the accompanying drawings and embodiments. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​described in these embodiments do not limit the scope of the present application.

[0039] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0040] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0041] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0042] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0043] like Figure 1 As shown, S1, performing channel optimization processing on the carrier image to obtain a sample image to adapt to the compression environment after uploading;

[0044] Furthermore, the specific content of the channel optimization processing of the carrier image in S1 is:

[0045] applying an additive distortion cost function to the carrier image and identifying a number of modifiable portions of the carrier image;

[0046] The visual impact caused by the modification in the modifiable part is quantified by an additive distortion cost function;

[0047] According to the result of the quantitative calculation, the initial distortion cost of the modifiable part of the carrier image is obtained;

[0048] The sample images are generated by mapping the calculated initial distortion cost onto the carrier image through color encoding.

[0049] S2, using STC (Syndrome-Trellis Codes) to embed ECC (Error Correction Code) encryption information in the sample to obtain an embedded image;

[0050] Furthermore, the expression of STC (Syndrome-Trellis Codes) in S2 is as follows:

[0051]

[0052] Ext = H·P(Y);

[0053] where Emb(·) and Ext(·) represent the embedding algorithm and extraction algorithm respectively, m is the embedded message, P(Y) represents the image modifiable component, H is the parity check matrix, and C(m) = {z|Hz = m} represents the complement corresponding to m.

[0054] Robust Adaptive Steganography (DMAS) based on dither modulation is the latest robust steganography algorithm under the framework of "robust domain selection + ECC + STC code". The algorithm mainly selects the intermediate frequency DCT coefficients as the embedding domain and constructs the compression-resistant domain in combination with dither modulation. DMAS can be applied to the spatial domain or the DCT domain. Robust Adaptive Steganography (GMAS) based on generalized dither modulation improves detection and compression elasticity by generalized dither modulation and extending the DMAS embedding domain. DMAS only changes the coefficients to the middle of the nearest interval, and GMAS considers the possibility of bidirectional changes and introduces ternary embedding and asymmetric distortion to hide the message. In order to obtain the quantized DCT coefficients from the received stego-image, the receiver uses the same quantization table as the embedding method. The encrypted data is then extracted by STC decoding. Finally, the accurate confidential information is obtained by RS decoding.

[0055] S3, modify the embedded image and splice the generated target image.

[0056] Furthermore, the specific steps of modifying the embedded image in S3 to generate the target image are:

[0057] The embedded image is divided into several 3×3 non-overlapping pixel blocks to generate the initial modification pattern;

[0058] Adjust the embedding direction within non-overlapping pixel blocks, traverse the non-zero elements, and generate 2 N embedding candidate patterns;

[0059] Calculate the embedding impact of each candidate pattern and sort them from small to large, and select the candidate pattern with the highest order as the embedding modification pattern;

[0060] The target image is generated by splicing non-overlapping pixel blocks according to the embedded modification pattern.

[0061] Further, the distortion cost is continuously adjusted when adjusting the embedding direction within the non-overlapping blocks and an asymmetric distortion cost adjustment strategy and an asymmetric distortion cost adjustment strategy are applied to the adjustment changes of the non-overlapping pixel blocks;

[0062] The asymmetric distortion cost adjustment strategy includes scaling the initial distortion cost according to the pixel modification direction.

[0063] There are various good distortion functions in the spatial domain, but they are almost all designed to guarantee the distortion of unit embedding changes. In this regard, the present application makes a simple but effective assumption that the spatial distortion cost is linearly proportional to the pixel modification amplitude.

[0064] Furthermore, the expression of the asymmetric distortion cost adjustment strategy is:

[0065]

[0066] Among them, the initial distortion cost of the “+1” and “-1” changes of each pixel is symmetrical, that is, ρ + =ρ - =ρ, d0 is the modified pixel x i,j of and, is the optimal change direction pixel x after the change i,j and the asymmetric distortion cost after the changes of “+1” and “-1”, α is the scaling factor (α>1), and the initial distortion cost ρ 0 Modified by "0" for pixels or DCT coefficients, defaults to 0.

[0067] This application experimentally studies the performance of this method. First, an experimental setup is proposed, and then the robustness and security of the proposed method are tested for JPEG compression.

[0068] The experimental dataset is BOSSBase 1.01, which contains 10,000 grayscale images in the format of 512 × 512. This method requires the use of an existing JPEG image distortion function to calculate the initial distortion. Any distortion function can be used here.

[0069] J-UNIWARD and UERD were used in the experiments. Both methods treat all coefficients as variable elements. The coding scheme is a ternary version of multi-layer STC. The larger the height h of the submatrix, the higher the security, but it also increases the computational complexity and reduces the redundancy due to error diffusion. In the experiment, h=10 was set. Similarly, preprocessing the message using ECC-based operations may sacrifice security to improve reliability. Due to performance limitations, previous steganography methods typically set the relative payload range of robust adaptive steganography to 0.05~0.15 bits / non-zero AC-DCT coefficient (bpnzac). In this application, this payload is called a low payload, and experiments are performed using a high payload of 0.1 to 0.5 (bpnzac).

[0070] When using OSS for image transmission, it will be recompressed with an unknown quality factor. The performance in this case can explain the ability of the method to resist recompression with an unknown quality factor. Therefore, its anti-recompression performance will be tested with different quality factors.

[0071] In MATLAB, imwrite is used to compress the images in BOSSBase 1.01 into cover images with quality factors of 70 and 80. Then various methods are used to obtain the stego images. The stego images with different quality factors are compressed again, the information is extracted, and the average information error rate is calculated. The results are shown in Tables 1 and 2. Table 1 is a comparison of the anti-JPEG recompression performance based on J-UNIWARD and UERD at QF=70, and Table 2 is a comparison of the anti-JPEG recompression performance based on J-UNIWARD and UERD at QF=80

[0072] like Figure 2 , Figure 3 , Figure 4 and Figure 5As shown in the figure, GMAS (Generalized Dither Modulation-basedrobust Adaptive Steganography) is an adaptive robust steganography algorithm based on generalized dither modulation. JCRIS (JPEG Compression-Resistant Information Steganography) is a steganography algorithm specially designed to resist JPEG compression. J-UNIWARD (JPEG Universal Wavelet Relative Distortion) is a steganography algorithm based on JPEG images. UERD (Universal Embedding Distortion for Recompression-resistant Data hiding) is a universal embedding distortion function. Propose refers to the method proposed in this paper.

[0073] DCTR is a feature extractor based on discrete cosine transform residual, which is used to extract features from images for steganalysis. SRNet is a classifier based on deep neural network, which is specially used for image steganalysis.

[0074] d) GMAS DCTR and GMAS SRNet refer to the steganalytic images generated using the GMAS method and detected by the DCTR and SRNet classifiers, respectively.

[0075] JCRIS is another steganographic method that is compared with the GMAS method. Similarly, DCTR and SRNet are classifiers used to detect steganographic images generated by the JCRIS method.

[0076] JCRIS DCTR and JCRIS SRNet represent the detection of steganographic images generated by the JCRIS method using the DCTR and SRNet classifiers, respectively.

[0077] , usually represents the average error rate (Average Error Rate). In steganalysis, this is an important indicator used to measure the security of the steganographic method or the error probability that the steganographic image is identified as "non-steganographic" by the detection classifier. It can be seen from the data that the extraction success rate of the method of the present application has been improved and the robustness has also been significantly improved. This is mainly because the method utilizes the robustness of the image itself to JPEG compression and uses pixel changes and error correction characteristics suitable for the image. As the workload increases, the error rate of message extraction is also steadily increasing.

[0078] Table 1

[0079]

[0080] The cover image used in steganalysis is the original image compressed by BOSSBase 1.01, and the stego-image is the corresponding image generated by various methods. In order to test the security performance, a built-in FLD classifier is used, which is trained by DCTR features and SRNet. The input of the built-in classifier is the features extracted from the mask and the stego-image. The output is the classification error rate, which is the average of the false alarm rate and the false alarm rate of more than 10 executions when using a 5000 / 5000 database split. The security of the method is evaluated using the classification error rate and the test error rate. The present invention calls them

[0081] Table 2

[0082]

[0083] Since online social software (OSS) usually performs lossy operations on uploaded images. Steganography methods used for OSS must be robust and undetectable. However, previous methods require significant compromises in security to achieve robustness. The present invention proposes a new method for replacing masked image coefficients based on the optimal steganography method for masked images after channel processing. Experimental results show that the method proposed in the present invention significantly improves its compressive strength and security.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A block-by-block reorganization JPEG steganography method based on channel optimization, It is characterized in that The following steps are involved: S1, performing channel optimization processing on the carrier image to obtain a sample image to adapt to the compression environment after uploading; S2, using STC (Syndrome-Trellis Codes) to embed ECC (Error Correction Code) encryption information in the sample to obtain an embedded image; S3, modify the embedded image and splice the generated target image.

2. The block-by-block reorganization JPEG steganography method based on channel optimization according to claim 1, It is characterized in that The specific content of channel optimization processing of the carrier image in S1 is: applying an additive distortion cost function to the carrier image and identifying a number of modifiable portions of the carrier image; The visual impact caused by the modification in the modifiable part is quantified by an additive distortion cost function; According to the result of the quantitative calculation, the initial distortion cost of the modifiable part of the carrier image is obtained; The sample images are generated by mapping the calculated initial distortion cost onto the carrier image through color encoding.

3. The block-by-block reorganization JPEG steganography method based on channel optimization according to claim 1, It is characterized in that The expression of STC (Syndrome-Trellis Codes) in S2 is as follows: Ext = H·P(Y); Wherein, Emb(·) and Ext(·) represent the embedding algorithm and the extraction algorithm respectively, m is the embedded message, P(Y) represents the image modifiable component, H is the parity check matrix, and C(m)=z|Hz=m} represents the complement corresponding to m.

4. The block-by-block reorganization JPEG steganography method based on channel optimization according to claim 3, It is characterized in that The specific steps for modifying the embedded image in S3 to generate the target image are: The embedded image is divided into several 3×3 non-overlapping pixel blocks to generate an initial modification pattern; Adjust the embedding direction within non-overlapping pixel blocks, traverse the non-zero elements, and generate 2 N embedding candidate patterns; Calculate the embedding impact of each candidate pattern and sort them from small to large, and select the candidate pattern with the highest order as the embedding modification pattern; The target image is generated by splicing non-overlapping pixel blocks according to the embedded modification pattern.

5. The block-by-block reorganization JPEG steganography method based on channel optimization according to claim 4, It is characterized in that Continuously adjusting the distortion cost when adjusting the embedding direction within the non-overlapping blocks and applying an asymmetric distortion cost adjustment strategy and an asymmetric distortion cost adjustment strategy to the adjustment changes of the non-overlapping pixel blocks; The asymmetric distortion cost adjustment strategy includes scaling the initial distortion cost according to the pixel modification direction.

6. The block-by-block reorganization JPEG steganography method based on channel optimization according to claim 5, It is characterized in that The expression of asymmetric distortion cost adjustment strategy is: Among them, the initial distortion cost of "+1" and "-1" changes of each pixel is symmetrical, that is, ρ + =ρ - =ρ,d 0 is the modified pixel x i,j of and, is the optimal change direction pixel x after the change i,j and the asymmetric distortion cost after changing to "+1" and "-1", α is the scaling factor (α>1), and the initial distortion cost ρ 0 Modified by "0" for pixels or DCT coefficients, defaults to 0.

Images