JPEG (Joint Photographic Experts Group) embedding optimization method based on channel overflow processing

Abstract

The invention discloses a JPEG (Joint Photographic Experts Group) embedding optimization method based on channel overflow processing, which relates to the field of image processing, and comprises the following steps of: S1, carrying out scaling overflow processing on a cover through a scale formula to obtain a robust cover, S2, calculating a distortion value of the robust cover by utilizing a J-UNIWARD distortion function and calculating asymmetric distortion, obtaining one-time modification cost through the asymmetric distortion, and carrying out one-time modification on the one-time modification cost through the asymmetric distortion. Modifying according to the primary modification cost; s3, performing data compression and information hiding by adopting STCs coding to obtain a temporary image; and S4, performing JPEG compression on the temporary image and extracting an error rate according to compression information, if the error rate meets a threshold value, generating a final image, and if the error rate does not meet the threshold value, adjusting the error correction capability and repeating S3 and S4 until the final image is generated. The method has good robustness on JPEG pictures uploaded and downloaded by a public social platform, and solves the problem of picture distortion to a certain extent.

Description

Technical Field

[0001] The invention relates to the field of image processing, in particular to a JPEG embedding optimization method based on channel overflow processing. Background Art

[0002] The JPEG embedding optimization method based on channel overflow processing is a technical means specifically for the data overflow problem that occurs during digital image compression. JPEG (Joint Photographic Experts Group) is a compression standard widely used in digital image storage and transmission. Its purpose is to reduce the file size while maintaining high image quality.

[0003] In the JPEG image compression process, the image is converted from the spatial domain to the frequency domain through discrete cosine transform (DCT), and then the obtained DCT coefficients are quantized. The quantization process involves a large number of numerical operations, which may cause data overflow, especially when processing images with a large dynamic range. This overflow not only affects the image quality, but may also cause serious data loss.

[0004] In order to solve this problem, an optimization method based on channel overflow processing is proposed. The core of this method is to avoid data loss and image quality degradation by performing special processing on color channels that may overflow. Specifically, this method can be divided into several key steps: first, monitor each color channel in the JPEG encoding process to detect whether there is an overflow risk; second, apply specific algorithm adjustments to those channels that may cause overflow, such as adjusting the quantization step size or reducing the coefficient value through mathematical methods to ensure that it is within the allowable numerical range; finally, re-encode these adjusted channels to generate the final JPEG image.

[0005] The advantage of this technology is that it can effectively control the data overflow problem in the JPEG compression process while maintaining the image quality as much as possible. This is not only crucial for applications that require high-fidelity images (such as medical imaging, professional photography, etc.), but also has a direct impact on the higher quality image experience that ordinary consumers can get when taking photos with digital cameras or smartphones.

[0006] The JPEG embedding optimization method based on channel overflow processing effectively avoids data overflow problems and ensures the quality of output images by finely managing each color channel in the image compression process. This technology not only improves the applicability and reliability of the JPEG standard, but also provides important reference and guidance for the development of future image compression technology. Summary of the invention

[0007] To solve the above problems, the present application provides a JPEG embedding optimization method based on channel overflow processing, with the goal of improving robustness, aiming to eliminate the impact of spatial overflow of the overlay image during the embedding process, thereby achieving the purpose of improving anti-steganography.

[0008] In the present application, a JPEG embedding optimization method based on channel overflow processing is provided, comprising the following steps:

[0009] S1, scaling overflow processing is performed on the cover through a scaling formula to obtain a robust cover;

[0010] S2. Calculate the distortion value of the robust cover using the J-UNIWARD distortion function and calculate the asymmetric distortion, obtain a modification cost through the asymmetric distortion, and modify the robust cover according to the modification cost;

[0011] S3, using STCs coding to perform data compression and information hiding to obtain a temporary image;

[0012] S4, perform JPEG compression on the temporary image and extract the error rate based on the compression information. If the error rate meets the threshold, the final image is generated. If not, the error correction capability is adjusted and S3 and S4 are repeated until the final image is generated.

[0013] Preferably, the scaling formula in S1 is expressed as follows:

[0014] s′ i,j =α·s i,j ,i,j∈{0,1,…,7},

[0015]

[0016] where s′ i,j represents the scaled spatial value, α is a parameter. The size of α depends on the spatial overflow degree of the block, where s |max| Represents the spatial value with the largest absolute value. α is calculated in this way so that the scaling block does not overflow.

[0017] Preferably, in S1, during the scaling overflow process of the cover using the scaling formula, the cover image is divided into 8×8 pixel blocks, and discrete cosine transform (DCT) is performed on the pixel blocks to generate 64 DCT coefficients.

[0018] Preferably, the specific content of adjusting the error correction capability in S4 is:

[0019] Obtain secret information, encode the secret information using RS code, and obtain encoded information;

[0020] Load the coded information onto the temporary image compressed by JPEG to obtain the encrypted image;

[0021] If the secret information can still be completely extracted from the cryptanalyzed image after being compressed in the analog channel, the coded information is embedded into the robust cover sequence, a steganographic sequence is generated to obtain a steganographic image, and the steganographic image is directly output as the final image;

[0022] Set the error correction capability of RS coding to t=1. If the secret information cannot be completely extracted from the encrypted image after being compressed in the analog channel, set t=t+1 until the secret information can still be completely extracted from the encrypted image after being compressed in the analog channel.

[0023] If the error correction code threshold is reached and the secret information cannot be completely extracted from the encrypted image after being compressed in the analog channel, the embedding domain is adjusted and S3 and S4 are repeated;

[0024] Preferably, the specific content of adjusting the embedded domain is:

[0025] Using the jitter modulation algorithm, sample cover sequences and sample cover modification distances are extracted from the robust cover, and the secondary modification cost is calculated;

[0026] Embedding domain adjustment is performed according to sample cover sequence and sample cover modification distance;

[0027] Generate a stego-image based on the stego-sequence and the sample cover modification distance.

[0028] Preferably, the specific content of the jitter modulation algorithm in S3 is:

[0029] Perform DCT transformation on the pixels of the sample image to obtain a number of DCT coefficients, and select a number of DCT coefficients as carriers;

[0030] The carrier is quantized and rounded with △ as the quantization step, and the carrier is divided into two categories A and B according to the parity of the quantization rounding result.

[0031] In summary, the JPEG embedding optimization method based on channel overflow processing of the present invention preprocesses the cover image to prevent overflow, optimizes the fixed embedding domain of GMAS, and selects the appropriate error correction code length for different embedding domains compared to traditional image processing technology. The performance under various channel compression settings is evaluated through a large number of experiments with CCPEV and DCTR and effective steganalysis. The experimental results show that the method mentioned in this application is superior to GMAS in terms of robustness and anti-steganography.

[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 an overall framework diagram of a JPEG embedding optimization method based on channel overflow processing of the present invention;

[0034] Figure 2 A comparison diagram of the embedded domain of GMAS and the embedded domain of the present invention;

[0035] Figure 3 This is a comparison chart of the robustness of different algorithms when QF=85 of the present invention;

[0036] Figure 4 It is the robustness diagram of the three algorithms of the present invention;

[0037] Figure 5 It is a security diagram of the algorithm using CCPEV as the steganographic feature of the present invention;

[0038] Figure 6 It is a security diagram of the algorithm using DCTR as a steganographic feature of the present invention;

[0039] Figure 7 This is a graph showing the robustness and anti-steganalysis performance of different embedding domains of a JPEG embedding optimization method based on channel overflow processing of the present invention. DETAILED DESCRIPTION

[0040] 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.

[0041] 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.

[0042] 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.

[0043] 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.

[0044] 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.

[0045] like Figure 1 As shown, the present invention provides a JPEG embedding optimization method based on channel overflow processing, which is specifically as follows and includes the following steps:

[0046] S1, scaling overflow processing is performed on the cover through a scaling formula to obtain a robust cover;

[0047] Furthermore, the expression of the scaling formula in S1 is as follows:

[0048] s′ i,j =α·s i,j ,i,j∈{0,1,…,7},

[0049]

[0050] where s′ i,j represents the scaled spatial value, α is a parameter. The size of α depends on the spatial overflow degree of the block, where s |max| Represents the spatial value with the largest absolute value. α is calculated in this way so that the scaling block does not overflow.

[0051] Furthermore, in S1, during the scaling overflow process of the cover using the scaling formula, the cover image is divided into 8×8 pixel blocks, and discrete cosine transform (DCT) is performed on the pixel blocks to generate 64 DCT coefficients.

[0052] S2. Use the J-UNIWARD distortion function to calculate the distortion value of the robust cover and calculate the asymmetric distortion. Get a modification cost through the asymmetric distortion. Modify the robust cover according to the modification cost. The modification cost is to further optimize and adjust the original distortion value in order to better adapt to specific application scenarios or needs.

[0053] S3, using STCs coding to perform data compression and information hiding to obtain a temporary image;

[0054] S4, perform JPEG compression on the temporary image and extract the error rate based on the compression information. If the error rate meets the threshold, the final image is generated. If not, the error correction capability is adjusted and S3 and S4 are repeated until the final image is generated.

[0055] Furthermore, the specific content of adjusting the error correction capability in S4 is:

[0056] Get secret information, use RS code to encode the secret information, and get the encoded information. The secret information is the information to be hidden, which can be a piece of text. ;

[0057] Load the coded information onto the temporary image compressed by JPEG to obtain the encrypted image;

[0058] If the secret information can still be completely extracted from the cryptanalyzed image after being compressed in the analog channel, the coded information is embedded into the robust cover sequence, a steganographic sequence is generated to obtain a steganographic image, and the steganographic image is directly output as the final image;

[0059] Set the error correction capability of RS coding to t=1. If the secret information cannot be completely extracted from the encrypted image after being compressed in the analog channel, set t=t+1 until the secret information can still be completely extracted from the encrypted image after being compressed in the analog channel.

[0060] If the error correction code threshold is reached and the secret information cannot be completely extracted from the encrypted image after being compressed in the analog channel, the embedding domain is adjusted and S3 and S4 are repeated;

[0061] Furthermore, the specific content of the embedded domain is adjusted as follows:

[0062] Using the jitter modulation algorithm, the sample cover sequence and the sample cover modification distance are extracted from the robust cover, and the secondary modification cost is calculated. The secondary modification cost is the corresponding distortion when a part of the temporary image is modified.

[0063] Embedding domain adjustment is performed according to sample cover sequence and sample cover modification distance;

[0064] Generate a stego-image based on the stego-sequence and the sample cover modification distance.

[0065] Furthermore, the specific content of the jitter modulation algorithm is:

[0066] Perform DCT transformation on the pixels of the sample image to obtain a number of DCT coefficients, and select a number of DCT coefficients as carriers;

[0067] The carrier is quantized and rounded with △ as the quantization step, and the carrier is divided into two categories A and B according to the parity of the quantization rounding result.

[0068] The dataset used in the present invention is BOSSBase 1.01, which contains 10,000 grayscale 512×512 images, 2000 of which are randomly selected. The original images are JPEG compressed. The algorithm based on jitter modulation needs to select a cover with a relatively low QF to achieve robustness. Therefore, a quality factor of 65 is selected as the cover image. Previous steganography algorithms usually set the relative payload of robust adaptive steganography in the range of 0.05 to 0.15 bits / non-zero AC DCT coefficient (bpnzac). Due to its limited performance, the payload is relatively low.

[0069] In the present invention, a high payload of 0.1 to 0.5 (bpnzac) is used for experiments, and the relative payload is payload=n m / n nzac , where n m is the length of the original embedded message, not the length of the message encoded by the RS code, n nzac is the number of non-zero AC DCT coefficients of the original cover image, not the robust cover after de-overflow processing. The extraction error rate Rerror =n error / n m , where n error The quality factors of the overlay image and channel JPEG compression are denoted by Q cover and Q channel Two effective steganalysis feature sets CCPEV and DCTR are selected for steganalysis of JPEG images. The security parameter h of STCs is set to 10.

[0070] The detector with default settings is trained as a binary classifier using the FLD set. The security performance of the steganographic method is evaluated by the detection error rate on the test set, which is expressed as Where P FA and P MD are the false alarm probability and missed detection probability respectively. The final security is determined by the average error rate to limit, and larger This means stronger anti-steganographic performance. When evaluating the security of the algorithm, the image before preprocessing is used as the carrier and the image after steganography is used as the stegographic information for steganalysis.

[0071] The robustness of JPEG compression under different quality factors is compared. cover =65, Q channel =85 and Q channel =95 when the effective load is 0.1~0.5. The average extraction error rate of the proposed scheme, GMAS and ROAST-OS proposed by Zeng et al. is calculated. The proposed scheme is better than GMAS and ROAST-OS. The method has stronger robustness under various loads. This is because the image overflow is optimized and a more reasonable embedding method is designed. The experimental results are shown in Figure 3 and 4 As shown, GMAS (Generalized Dither Modulation-based robust Adaptive Steganography): GMAS is an adaptive robust steganography algorithm based on generalized dither modulation. JCRIS: JCRIS (JPEG Compression-Resistant Information Steganography) is a steganography algorithm designed specifically to resist JPEG compression. J-UNIWARD (JPEG Universal Wavelet Relative Distortion): J-UNIWARD is a steganography algorithm based on JPEG images. ROAST-OS refers to the performance evaluation used to analyze steganography or other technologies. Propose refers to the method proposed in this paper.

[0072] Preprocessing enhances the stability of DCT coefficients by eliminating spatial overflow in the covered image. This operation has a significant impact on security and robustness.

[0073] In addition, the preprocessing method for removing overflow proposed by the present invention is difficult to detect by the previous JPEG recompression detection method. The method has high robustness, mainly because it solves the coefficient overflow problem encountered in the image JPEG compression process.

[0074] like Figure 2 As shown, the present application utilizes the embedding domain suitable for the image and the error correction capability to actively and conveniently generate a robust overlay image. As the payload increases, the error rate in message extraction increases steadily. GMAS encrypts the secret message using ternary STCs and an error correction algorithm. As the payload increases, the probability that the number of errors in the decoded secret message exceeds the error correction capability also increases.

[0075] The average detection error rate The proposed algorithm is compared with the recent robust steganography methods. The security of the proposed method is compared with GMAS and ROAST-OS, and CCPEV and DCTR are used to demonstrate the anti-steganalysis performance of different algorithms. The experimental results are shown in Figure 5 and 6 As shown. Security and robustness are difficult to balance, so improving robustness usually requires sacrificing some aspects of security. When steganalysis is performed using the feature CCPEV, the difference in security is negligible, while when the feature DCTR is used, the difference in security is even negligible. Experimental results show that this method effectively improves the robustness of the image. This is achieved by preferentially embedding information in low-frequency areas, which have higher resistance to steganalysis. This method reduces overflow during image quantization while ensuring image robustness, and imposes a threshold limit on the length of the error correction code. Therefore, the generated stego-image does not show obvious distortion, thereby enhancing resistance to steganalysis.

[0076] When the load is small, the security of the algorithm depends largely on the preprocessing method. When the load is large, the modifications caused by embedding will gradually take effect. The method proposed in the present invention performs adaptive embedding based on overflow removal. Therefore, it has good performance when the load is small. As the load increases, the performance gradually decreases, but compared with other methods, it still has a good performance advantage.

[0077] The performance of the proposed algorithm is tested using Facebook as an example. Based on the previous pattern of Facebook choosing a heavier compression quality factor, it is obvious that the quality factor of most images on Facebook is less than 72. However, too low a quality factor will degrade the image quality. Therefore, I choose 60 images with hidden information (randomly selected from BOSSbase 1.01, Qcover = 60, 65, 70, each containing 20 pictures) as examples. In this experiment, the algorithm is applied to Facebook, and the message error rate is shown in Table 1. The results show that the algorithm can achieve error-free transmission when the coverage quality factor is 60, and it also achieves error-free transmission well when the coverage quality factor is 70. As expected, the solution can be effectively applied in the real world.

[0078] Table 1 The proposed method and GMAS and N success Compared with Facebook, represents the average error rate. It is usually used to measure the average proportion of errors generated by an algorithm or model in a series of tests. N_success represents the number of successes, that is, the number of times the expected results are successfully achieved in a series of experiments or tests. This indicator is usually used to measure the effectiveness of an algorithm or method. 60 images were randomly selected from BOSSbase 1.01, with statistical errors = 60, 65, 70, each containing 20 images of 0.1bpnzac

[0079] Table 1

[0080]

[0081] In order to verify the performance of the improved adaptive embedding domain method, 2000 images were randomly selected from BOSSbase 1.01, where Qcover = 75. The message embedding payload is 0.15, and the RS error correction capability is fixed to 8. In order to detect the steganographic image, the feature DCTR is used and Qchannel = 75 is set to simulate the channel JPEG compression. Figure 7 As shown in the figure, the embedding domain is gradually increased from low frequency to medium frequency. As the embedding domain shifts from low frequency to medium frequency, the robustness is enhanced. Since the coverage overflow is preprocessed, the loss in anti-steganalysis performance is minimal. The robustness of the embedding domain is significantly improved when the frequency is high. Therefore, the embedding area can be adaptively adjusted according to the robustness of the image. Due to the preprocessing operation, there is no need to consider the anti-steganalysis performance too much, and a balance between robustness and anti-steganalysis performance is achieved.

[0082] Based on the derivation of the influence of embedding in low-frequency areas, an adaptive embedding method resistant to JPEG compression is designed. Because the cover image is processed to eliminate overflow before embedding information, the capacity of embedded information and the ability to resist steganalysis are greatly enhanced. Combined with the embedding domain proposed by the present invention and the error correction code length designed for different embedding domains, the robustness and security are improved while ensuring the embedding capacity.

[0083] 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 JPEG embedding optimization method based on channel overflow processing, characterized in that: The following steps are involved: S1, scaling overflow processing is performed on the cover through a scaling formula to obtain a robust cover; S2. Calculate the distortion value of the robust cover using the J-UNIWARD distortion function and calculate the asymmetric distortion, obtain a modification cost through the asymmetric distortion, and modify the robust cover according to the modification cost; S3, using STCs coding to perform data compression and information hiding to obtain a temporary image; S4, perform JPEG compression on the temporary image and extract the error rate based on the compression information. If the error rate meets the threshold, the final image is generated. If not, the error correction capability is adjusted and S3 and S4 are repeated until the final image is generated.

2. The JPEG embedding optimization method based on channel overflow processing according to claim 1, characterized in that: The expression of the scaling formula in S1 is as follows: s i ' ,j =α·s i,j ,i,j∈{0,1,|,7}, where s' i,j represents the scaled spatial value, α is a parameter. The size of α depends on the spatial overflow degree of the block, where s |max| Represents the spatial value with the largest absolute value. α is calculated in this way so that the scaling block does not overflow.

3. The JPEG embedding optimization method based on channel overflow processing according to claim 1, characterized in that: In S1, during the scaling overflow process of the cover using the scaling formula, the cover image is divided into 8×8 pixel blocks, and discrete cosine transform (DCT) is performed on the pixel blocks to generate 64 DCT coefficients.

4. The JPEG embedding optimization method based on channel overflow processing according to claim 1, characterized in that: The specific contents of adjusting the error correction capability in S4 are: Obtain secret information, encode the secret information using RS code, and obtain encoded information; Load the coded information onto the temporary image compressed by JPEG to obtain the encrypted image; If the secret information can still be completely extracted from the cryptanalyzed image after being compressed in the analog channel, the coded information is embedded into the robust cover sequence, a steganographic sequence is generated to obtain a steganographic image, and the steganographic image is directly output as the final image; Set the error correction capability of RS coding to t=1. If the secret information cannot be completely extracted from the encrypted image after being compressed in the analog channel, set t=t+1 until the secret information can still be completely extracted from the encrypted image after being compressed in the analog channel. If the error correction code threshold is reached and the secret information cannot be completely extracted from the encrypted image after being compressed in the analog channel, the embedding domain is adjusted and S3 and S4 are repeated.

5. The JPEG embedding optimization method based on channel overflow processing according to claim 4 is characterized in that: Adjust the specific content of the embedded domain as follows: Using the jitter modulation algorithm, sample cover sequences and sample cover modification distances are extracted from the robust cover, and the secondary modification cost is calculated; Embedding domain adjustment is performed according to sample cover sequence and sample cover modification distance; Generate a stego-image based on the stego-sequence and the sample cover modification distance.

6. The JPEG embedding optimization method based on channel overflow processing according to claim 5, characterized in that: The specific contents of the jitter modulation algorithm are: Perform DCT transformation on the pixels of the sample image to obtain a number of DCT coefficients, and select a number of DCT coefficients as carriers; The carrier is quantized and rounded with △ as the quantization step, and the carrier is divided into two categories A and B according to the parity of the quantization rounding result.

Images