JPEG image reversible data hiding method based on point distortion estimation

Abstract

The invention discloses a joint photographic experts group (JPEG) image reversible data hiding method based on point distortion estimation, and belongs to the technical field of information hiding. The method mainly comprises the steps of 1, a preprocessing process; 2, embedding secret information; according to the method, a point distortion estimation model considering block smoothness, quantization step size and Zig-zag sorting positions is constructed, then AC coefficients are arranged in an ascending order according to point distortion estimation values, and better image quality can be obtained due to the fact that the AC coefficients with small distortion are embedded preferentially. Secondly, a principle of'meeting + / -1 pairing 'is designed, and a + / -1 coefficient is fully utilized to improve the embedding capacity. And finally, embedding the secret data into the AC coefficient pair by using a mapping strategy combining a one-dimensional histogram and a two-dimensional histogram. The algorithm provided by the invention can provide better visual image quality, meanwhile, the increase of the file size is small, and the effect of time complexity is very ideal.

Description

Technical Field

[0001] The invention belongs to the technical field of information hiding, and in particular relates to a JPEG image reversible data hiding method based on point distortion estimation. Background Art

[0002] With the development of the Internet and portable digital products, multimedia data such as images and videos have become the main medium for information transmission in the Internet. More and more users choose to use public platforms in the Internet such as Weibo, WeChat and cloud disk to store personal data, which brings certain security threats to the privacy data of users. Information hiding technology is an important data protection technology. It uses the insensitivity of the human visual system and the redundancy of the image itself to embed the information to be protected into another public carrier. The carrier after the embedded information is visually indistinguishable from the original carrier, thus ensuring the security of the secret information. However, the embedding process will change the pixel value of the original carrier, resulting in permanent distortion after the secret data is extracted, which is unacceptable for some sensitive application fields such as medical and military. Therefore, reversible data hiding has been widely studied. It can accurately extract secret data and restore the original carrier image without distortion to meet the needs of these image sensitive fields. The Joint Photographic experts group (JPEG) is one of the most widely used image formats in the Internet. However, JPEG images have problems such as low redundancy, limited embedding capacity, and the modification of arbitrary data in the compression domain easily causes large visual distortion. Therefore, information hiding on JPEG images is more challenging.

[0003] In 2015, Huang et al. (Huang Fangjun, Qu Xiaochao, Kim HJ, et al. Reversible data hiding in JPEG images [J]. IEEE Transactions on Circuits and Systems for Video Technology, 2015, 26 (9): 1610-1621.) proposed a JPEG information hiding method based on one-dimensional DCT coefficient histogram shift. This method embeds secret information into AC coefficients with an absolute value of 1, while the DC coefficient and the zero coefficient remain unchanged, and the remaining AC coefficients are shifted to make redundant space for the secret information. At the same time, this method proposes to give priority to DCT blocks with more zero coefficients for embedding, which increases the embedding capacity while maintaining the file size. Later, Hou et al. (Hou Dongdong, Wang Haoqian, Zhang Weiming, et al. Reversible data hiding in JPEG image based on DCT frequency and block selection [J]. Signal Processing, 2018, 148: 41-47.) improved the above method and introduced a new block sorting and frequency selection strategy. The AC coefficients at the frequency position with less distortion in the DCT coefficient block were embedded to improve the visual quality of the image. Li et al. proposed a JPEG information hiding algorithm based on the modification of the two-dimensional histogram of quantized DCT coefficients. First, two adjacent non-zero AC coefficients are paired to construct a two-dimensional histogram; then, the coefficient pairs are modified using the designed two-dimensional reversible mapping rules to embed secret information. Although the algorithm improves image quality and reduces file increment, the amplitude relationship between coefficients is not considered in the coefficient pairing process, so the optimal rate-distortion performance cannot be obtained. Therefore, Wu et al. (WU Taoyu, HUANG Fangjun. Adaptive reversible data hiding in JPEG images based on coefficient pairing [J]. Journal of Software, 2022, 33 (02): 725 + 728-739.) proposed an adaptive JPEG information hiding algorithm based on coefficient pairing, which uses small AC coefficients to pair two by two to construct a two-dimensional reversible mapping rule, and combines the adaptive frequency strategy to select coefficients of different frequencies in each sub-block to embed secret information. Experimental results show that compared with similar algorithms, this algorithm has obvious advantages in terms of visual quality and file increment.However, although the above solutions have achieved good visual quality, they still cannot meet the application requirements in the existing environment in terms of embedding capacity and file increment performance. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a JPEG image reversible data hiding method based on point distortion estimation. Compared with other JPEG image reversible data hiding methods, the method of the present invention can provide better visual image quality guarantee, relatively small file size increment, and more ideal running time complexity.

[0005] The technical solution protected by the present invention is: a JPEG image reversible data hiding method based on point distortion estimation, which is specifically carried out according to the following steps:

[0006] Step S1, pre-processing process, specifically includes the following steps:

[0007] Step S11, read in a carrier image I of size M×N, and divide it into non-overlapping image blocks of size 8×8;

[0008] Step S12: Perform DCT transformation on each 8×8 image block to convert the original image from the spatial domain to the frequency domain:

[0009] Step S13, using an 8×8 quantization table to quantize the obtained DCT coefficients to obtain quantized DCT coefficients d(a, b);

[0010] Step S14: Scan each 8×8 quantized DCT coefficient block in zig-zag order to obtain a coefficient sequence where k∈{1,2,...,N} represents the kth quantized DCT coefficient block, are DC coefficients, and the remaining 63 are AC coefficients;

[0011] Step S15: Select the first k non-continuous zero AC coefficient sequences in each coefficient block, denoted as:

[0012] where ξ s (C k ) represents the number of non-consecutive zero coefficients;

[0013] Step S16: Calculate the distortion of all non-zero AC coefficients according to the point distortion estimation model, as shown in formula (3):

[0014]

[0015] in, The point distortion estimate of the non-zero AC coefficient at the zig-zag position l in the kth DCT block, Q(l) is the quantization step size at the zig-zag position l, and α and β are adjustment factors;

[0016] Step S17: Distortion estimation value of each AC coefficient Arrange in ascending order to form a new sequence

[0017] where δ:{(1,1),..(k,l),...(4096,63)}→{δ(1),δ(2),...,δ(θ),...,δ(4096×63)} is a one-to-one mapping such that D δ(1) ≤D δ(2) ≤...≤D δ(θ) ≤...≤D δ(4096*63) ; When D δ(p) =D δ(q) When p<q, δ(p)<δ(q);

[0018] Step S18: Based on the point distortion estimation value sequence Φ={D δ(θ) |D δ(θ) ≠0} θ＝1to 4096×63 , and the corresponding AC coefficient sequence Ψ={d δ(θ) |d δ(θ) ≠0} θ＝1to 4096×63 , using the "pairing when encountering ±1" principle, pair ±1 with its adjacent non-zero AC coefficients to obtain the embeddable coefficient pair set DS, and the remaining unpaired AC coefficients are the movable coefficient set SS;

[0019] Step S2: The embedding process of secret information specifically includes the following steps:

[0020] Step S21, divide the embeddable coefficient pair set DS and the movable coefficient set SS into three categories:

[0021] Type A: coefficient pairs whose values ​​in DS are (-1, -1), (-1, 1), (1, -1), (1, 1);

[0022] Type B: coefficient pairs in DS other than type A;

[0023] Type C: coefficients in SS;

[0024] Step S22: for different types of coefficient pairs, use a mapping rule combining 2D and 1D;

[0025] Step S23: read the secret data in sequence, and select different embedding rules for different types of coefficient pairs:

[0026] Step S24, repeat the above third step until all given secret data are embedded;

[0027] Step S25, entropy encoding is performed on the quantized DCT coefficients embedded with the secret data to obtain a secret JPEG image;

[0028] Step 3: The process of extracting secret information includes the following steps:

[0029] Step S31, entropy decoding is performed on the encrypted JPEG image to obtain a quantized DCT coefficient block;

[0030] Step S32, according to the fourth, fifth, sixth and seventh steps in the preprocessing process, a coefficient set is obtained. When the AC coefficient value is ±1 and ±2, it will be paired with the next adjacent AC coefficient. The paired coefficient is placed in the DS set, and the unpaired coefficient is placed in the SS set.

[0031] Step S33, the AC coefficients in the SS set do not contain any embedded secret data. Therefore, when the AC coefficient is greater than 0, subtracting 1 can restore the original value, and conversely, adding 1 can restore the original value. However, for the coefficient pairs in the DS set, the secret data is extracted and the coefficients are restored in the same way as in step S23.

[0032] Furthermore, the specific formula of DCT transformation in step S12 is as follows:

[0033]

[0034] in, p(i, j) is the original image pixel value, P(a, b) represents the frequency domain coefficient value, and its frequency is {a, b};

[0035] Furthermore, the calculation formula of the quantized DCT coefficient d(a, b) in step S13 is as follows:

[0036]

[0037] Among them, Q(a,b) is the quantization step size at the corresponding frequency (a,b) in the quantization table, and d(a,b) is the quantized DCT coefficient.

[0038] Furthermore, the specific mapping rules of step S22 are as follows:

[0039] Type A: Can make horizontal, vertical and non-directional corrections according to specific circumstances;

[0040] Type B: Can be corrected in horizontal and vertical directions according to specific circumstances;

[0041] Type C: Horizontal correction.

[0042] Furthermore, in step S23, the embedding rules of different types of coefficients are as follows:

[0043] For type A, the embedding rules are divided into the following three cases:

[0044] When the secret data is 0, the coefficient pair does not change direction;

[0045] When the secret data is 1, the second secret data is read. When the second secret data is 0, the coefficient pair is corrected in the horizontal direction. When the second secret data is 1, the coefficient pair is corrected in the vertical direction.

[0046] For type B, the embedding rules are divided into the following two cases

[0047] When the secret data is 0, the coefficients are corrected horizontally;

[0048] When the secret data is 1, the coefficient is corrected in the vertical direction;

[0049] For type C, no secret data is embedded and only horizontal movement is performed to ensure reversibility.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] 1. The present invention fully considers the influence of three factors, namely block smoothness, frequency position selection and quantized DCT coefficients, on distortion. At the same time, the influence factor is used to dynamically adjust the influence of each factor on distortion. A point distortion estimation model is designed. By sorting the point distortion estimates, quantized DCT coefficients with small distortion are preferentially selected for embedding, thereby improving the visual quality of encrypted JPEG images.

[0052] 2. The present invention designs a new AC coefficient pairing strategy, implements the principle of "pairing when encountering ±1", makes full use of the ±1 coefficients for embedding, thereby improving the embedding capacity, while avoiding the disadvantage of embedding on the 0 coefficient resulting in an increase in file increments.

[0053] 3. The mapping method of the scheme of the present invention is simple in design, takes less running time, has low complexity and is easy to implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0055] Figure 1 Coefficient mapping rules for combining 1D and 2D.

[0056] Figure 2 Performance comparison of different images under QF=70.

[0057] Figure 3Average PSNR comparison of test images under fixed embedding capacity (QF=70, 80, 90).

[0058] Figure 4 Comparison of average file increments for test images at fixed embedding capacity (QF=70, 80, 90).

[0059] Figure 5 Average SSIM comparison for test images at fixed embedding capacity (QF=70, 80, 90).

[0060] Figure 6 Time complexity comparison of fixed embedding capacity under QF=70 for test images. DETAILED DESCRIPTION

[0061] In order to make the objects, features and advantages of the present invention clearly understood, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0062] We selected Matlab R2020b as the software platform to implement the design of the solution of the present invention. During the implementation process, a 512×512 standard test image "Boat" was selected as the carrier image.

[0063] The technical solution of the present invention consists of three parts: (I) preprocessing process; (II) embedding of secret information; and (III) extraction of secret information.

[0064] Ⅰ. The pretreatment process specifically includes the following steps:

[0065] In the first step, a 512×512 standard test image “Boat” is read as the carrier image I and divided into non-overlapping 8×8 image blocks;

[0066] In the second step, each 8×8 image block is subjected to DCT transformation using formula (1) to convert the original image from the spatial domain to the frequency domain:

[0067]

[0068] in, p(i, j) is the original image pixel value, P(a, b) represents the frequency domain coefficient value, and its frequency is {a, b};

[0069] The third step is to use an 8×8 quantization table to quantize the obtained DCT coefficients to obtain quantized DCT coefficients, as shown in formula (2):

[0070]

[0071] Among them, Q(a,b) is the quantization step size at the corresponding frequency (a,b) in the quantization table, and d(a,b) is the quantized DCT coefficient;

[0072] The fourth step is to scan each 8×8 quantized DCT coefficient block in zig-zag order to obtain the coefficient sequence where k∈{1,2,...,N} represents the kth quantized DCT coefficient block, are DC coefficients, and the remaining 63 are AC coefficients;

[0073] The fifth step is to select the first k non-continuous zero AC coefficient sequences in each coefficient block, denoted as where ξ s (C k ) represents the number of non-consecutive zero coefficients;

[0074] In the sixth step, the distortion of all non-zero AC coefficients is calculated according to the point distortion estimation model, as shown in formula (3):

[0075]

[0076] in, The point distortion estimate of the non-zero AC coefficient at the zig-zag position l in the kth DCT block, Q(l) is the quantization step size at the zig-zag position l, and α and β are adjustment factors;

[0077] Step 7: Estimate the point distortion of each AC coefficient Arrange in ascending order to form a new sequence Φ={D δ(θ) |D δ(θ) ≠0} θ＝1to 4096×63 , where δ:{(1,1),..(k,l),...(4096,63)}→{δ(1),δ(2),...,δ(θ),...,δ(4096×63)} is a one-to-one mapping such that D δ(1) ≤D δ(2) ≤...≤D δ(θ) ≤...≤D δ(4096*63) ; When D δ(p) =D δ(q) When p<q, δ(p)<δ(q);

[0078] Step 8: Based on the point distortion estimation value sequence Φ = {D δ(θ) |D δ(θ) ≠0} θ＝1to 4096×63 , and the corresponding AC coefficient sequence Ψ={d δ(θ) |d δ(θ) ≠0} θ＝1to 4096×63, using the "pairing when encountering ±1" principle, pair ±1 with its adjacent non-zero AC coefficients to obtain the embeddable coefficient pair set DS, and the remaining unpaired AC coefficients are the movable coefficient set SS;

[0079] II. The embedding process of secret information includes the following steps:

[0080] In the first step, the embeddable coefficient pair set DS and the movable coefficient set SS are divided into three categories:

[0081] Type A: coefficient pairs whose values ​​in DS are (-1, -1), (-1, 1), (1, -1), (1, 1);

[0082] Type B: coefficient pairs in DS other than type A;

[0083] Type C: coefficients in SS;

[0084] The second step is to use a mapping rule combining 2D and 1D for different types of coefficient pairs; (e.g. Figure 1 (shown)

[0085] (1) Type A: Horizontal, vertical and directional corrections can be made according to specific circumstances;

[0086] (2) Type B: Horizontal and vertical corrections can be made according to specific circumstances;

[0087] (3) Type C: horizontal correction;

[0088] The third step is to read the secret data in sequence and select different embedding rules for different types of coefficient pairs:

[0089] For type A, the embedding rules are divided into the following three cases:

[0090] (1) When the secret data is 0, the coefficient pair does not change direction;

[0091] (2) When the secret data is 1, the second secret data is read. When the second secret data is 0, the coefficient pair is corrected in the horizontal direction. When the second secret data is 1, the coefficient pair is corrected in the vertical direction.

[0092] For type B, the embedding rules are divided into the following two cases

[0093] (1) When the secret data is 0, the coefficient is corrected horizontally;

[0094] (2) When the secret data is 1, the coefficients are corrected in the vertical direction;

[0095] For type C, no secret data is embedded, and only horizontal movement is performed to ensure reversibility;

[0096] Step 4: Repeat step 3 until all given secret data are embedded.

[0097] The fifth step is to perform entropy coding on the quantized DCT coefficients embedded with secret data to obtain a secret JPEG image.

[0098] III. The extraction process of secret information is the reverse process of the embedding process, which specifically includes the following steps:

[0099] The first step is to entropy decode the encrypted JPEG image to obtain the quantized DCT coefficient block;

[0100] The second step is to obtain the coefficient set according to the fourth, fifth, sixth and seventh steps in the preprocessing process. When the AC coefficient value is ±1 and ±2, it will be paired with the next adjacent AC coefficient. The paired coefficient is placed in the DS set, and the unpaired coefficient is placed in the SS set.

[0101] In the third step, the AC coefficients in the SS set do not contain any embedded secret data. Therefore, when the AC coefficient is greater than 0, subtracting 1 can restore the original value, and vice versa, adding 1 can restore the original value. However, for the coefficient pairs in the DS set, the secret data is extracted and the coefficients are restored in the same way as in the third step;

[0102] In order to illustrate the superiority of the present invention over other schemes, we tested the performance of the present invention and other schemes on 6 standard test images of size 512×512 selected from the USC-SIPI dataset under the same experimental conditions. At the same time, the experiment also tested 1000 images selected from the image dataset BOSSbase and the dataset UCID. During the experiment, the secret data to be embedded was generated by a random generator. The experiment used three indicators: Peak Signal to Ratio (PSNR), structural similarity (SSIM) and increased filesize (IFS) to evaluate the performance of the algorithm.

[0103] Figure 2The PSNR, SSIM and IFS of 6 standard images of various algorithms under fixed embedding capacity when QF=70 are compared. It can be seen from the data in the table that compared with the four existing main algorithms, the present invention performs well in IFS. Although in a few cases, the present invention is slightly inferior to the comparison algorithm in PSNR or SSIM, this does not affect the overall advantage of the present invention. For example, when the embedding capacity is 16000 bits, the SSIM of the Elain image of the present invention is 0.9989, which is only 0.0001 lower than the algorithm of Wu et al. For the Peppers graph, the PSNR of the present invention is 40.73dB, which is 0.41dB lower than the algorithm of Hou et al.

[0104] Figure 3 , Figure 4 and Figure 5 The comparison between the embedding capacity and the average PSNR, average SSIM and average IFS of the five algorithms is shown when QF = 70, 80 and 90 respectively. Figure 3 It can be seen that when QF=70, the visual quality of the present invention is comparable to that of the algorithms of Hou et al., Weng et al., and Wu et al., and is superior to that of Huang et al. When QF=80, the present invention is slightly superior to the algorithms of Weng et al. and Wu et al.; when QF=90, the present invention has obvious advantages over other algorithms. Figure 4 It can be seen that the present invention is superior to other algorithms in terms of IFS.

[0105] At the same time, the experiment selected 1000 images from the dataset BOSSbase and the dataset UCID respectively, and performed performance tests at QF=70. At different embedding capacities, the average PSNR and average IFS of the present invention are better than those of other algorithms. Although the average SSIM of the present invention is comparable to that of the four most advanced algorithms, all SSIM values ​​exceed 0.9900, and the differences are very small, ensuring that there is no significant impact on the visual quality of the image. In addition, for the UCID dataset, at different fixed embedding capacities, the present invention has advantages in image visual quality, file increment, and structural similarity.

[0106] The above experiments prove the superiority of the present invention in maintaining image quality and minimizing file increment. However, in general, the goal of an excellent algorithm is to have the smallest time complexity. In order to verify the advantage of the present invention in terms of time complexity efficiency, the total running time of embedding 10,000 bits of secret data into 6 test images when QF=70 is recorded in the experiment. The results are summarized in Figure 6Obviously, the algorithm of Huang et al. has the shortest running time, while the algorithm of Weng et al. takes the longest time. The running time of the present invention is only 0.78 seconds slower than the algorithm of Huang et al., but it is significantly improved compared with the other three algorithms. Since the mapping method of the present invention is very simple, it is very competitive in terms of time complexity.

[0107] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A reversible data hiding method for JPEG images based on point distortion estimation, characterized in that: Follow these steps: Step S1, pre-processing process, specifically includes the following steps: Step S11, read in a carrier image I of size M×N, and divide it into non-overlapping image blocks of size 8×8; Step S12: Perform DCT transformation on each 8×8 image block to convert the original image from the spatial domain to the frequency domain: Step S13, using an 8×8 quantization table to quantize the obtained DCT coefficients to obtain quantized DCT coefficients d(a, b); Step S14: Scan each 8×8 quantized DCT coefficient block in zig-zag order to obtain a coefficient sequence where k∈{1,2,...,N} represents the kth quantized DCT coefficient block, are DC coefficients, and the remaining 63 are AC coefficients; Step S15: Select the first k non-continuous zero AC coefficient sequences in each coefficient block, denoted as: where ξ s (C k ) represents the number of non-consecutive zero coefficients; Step S16: Calculate the distortion of all non-zero AC coefficients according to the point distortion estimation model, as shown in formula (3): in, The point distortion estimate of the non-zero AC coefficient at the zig-zag position l in the k-th DCT block, Q(l) is the quantization step size at the zig-zag position l, and α and β are adjustment factors; Step S17: Distortion estimation value of each AC coefficient Arrange in ascending order to form a new sequence Φ={D δ(θ) |D δ(θ) ≠0} θ＝1 to 4096×63 ; where δ:{(1,1),..(k,l),...(4096,63)}→{δ(1),δ(2),...,δ(θ),...,δ(4096×63)} is a one-to-one mapping such that D δ(1) ≤D δ(2) ≤...≤D δ(θ) ≤...≤D δ(4096*63) When D δ(p) =D δ(q) When p<q, δ(p)<δ(q); Step S18: Based on the point distortion estimation value sequence Φ={D δ(θ) |D δ(θ) ≠0} θ＝1 to 4096×63 , and the corresponding AC coefficient sequence Ψ={d δ(θ) |d δ(θ) ≠0} θ＝1 to 4096×63 , using the "pairing when encountering ±1" principle, pair ±1 with its adjacent non-zero AC coefficients to obtain the embeddable coefficient pair set DS, and the remaining unpaired AC coefficients are the movable coefficient set SS; Step S2: The embedding process of secret information specifically includes the following steps: Step S21, divide the embeddable coefficient pair set DS and the movable coefficient set SS into three categories: Type A: coefficient pairs whose values ​​in DS are (-1, -1), (-1, 1), (1, -1), (1, 1); Type B: coefficient pairs in DS other than type A; Type C: coefficients in SS; Step S22: for different types of coefficient pairs, use a mapping rule combining 2D and 1D; Step S23: read the secret data in sequence, and select different embedding rules for different types of coefficient pairs: Step S24, repeat the above third step until all given secret data are embedded; Step S25, entropy encoding is performed on the quantized DCT coefficients embedded with the secret data to obtain a secret JPEG image; Step 3: The process of extracting secret information includes the following steps: Step S31, entropy decoding is performed on the encrypted JPEG image to obtain a quantized DCT coefficient block; Step S32, according to the fourth, fifth, sixth and seventh steps in the preprocessing process, a coefficient set is obtained. When the AC coefficient value is ±1 and ±2, it will be paired with the next adjacent AC coefficient. The paired coefficient is placed in the DS set, and the unpaired coefficient is placed in the SS set. Step S33, the AC coefficients in the SS set do not contain any embedded secret data. Therefore, when the AC coefficient is greater than 0, subtracting 1 can restore the original value, and conversely, adding 1 can restore the original value. However, for the coefficient pairs in the DS set, the secret data is extracted and the coefficients are restored in the same way as in step S23.

2. The method for reversible data hiding in JPEG images based on point distortion estimation according to claim 1, characterized in that: The specific formula of DCT transformation in step S12 is as follows: in, p(i, j) is the original image pixel value, P(a, b) represents the frequency domain coefficient value, and its frequency is {a, b}; 3. The method for reversible data hiding in JPEG images based on point distortion estimation according to claim 2, characterized in that: The calculation formula of the quantized DCT coefficient d(a, b) in step S13 is as follows: Among them, Q(a,b) is the quantization step size at the corresponding frequency (a,b) in the quantization table, and d(a,b) is the quantized DCT coefficient.

4. The method for reversible data hiding in JPEG images based on point distortion estimation according to claim 1, characterized in that: The specific mapping rules of step S22 are as follows: Type A: Can make horizontal, vertical and non-directional corrections according to specific circumstances; Type B: Can be corrected in horizontal and vertical directions according to specific circumstances; Type C: Horizontal correction.

5. The method for reversible data hiding in JPEG images based on point distortion estimation according to claim 4, characterized in that: In step S23, the embedding rules of different types of coefficients are as follows: For type A, the embedding rules are divided into the following three cases: When the secret data is 0, the coefficient pair does not change direction; When the secret data is 1, the second secret data is read. When the second secret data is 0, the coefficient pair is corrected in the horizontal direction. When the second secret data is 1, the coefficient pair is corrected in the vertical direction. For type B, the embedding rules are divided into the following two cases When the secret data is 0, the coefficients are corrected horizontally; When the secret data is 1, the coefficient is corrected in the vertical direction; For type C, no secret data is embedded and only horizontal movement is performed to ensure reversibility.

Images