Wavelet transform domain dual watermarking method based on image interpolation

Abstract

The invention relates to the technical field of image processing, and discloses a wavelet transform domain dual watermarking method based on image interpolation, which comprises the following steps of: firstly, improving embedding capacity by utilizing improved image interpolation; and then a watermark is ingeniously embedded in a low-frequency region of the original image after three-level transformation by using multilayer decomposition and local characteristics of wavelet transformation through a calculation method of changing a wavelet coefficient to realize a protection effect on image copyright. And secondly, carrying out three-level wavelet transformation on the carrier image embedded with the watermark on the basis, extracting coefficient correlation of a low-frequency part as an authentication watermark, and embedding the authentication watermark into a medium-high-frequency part of the two-level transformation to realize a protection function of image content and successfully locate a tampered position. Compared with the prior art, the method has high robustness and invisibility, successfully solves the problem that most watermark algorithms are single in function, and achieves balance between the robustness and the invisibility.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a wavelet transform domain double watermarking method based on image interpolation. Background Art

[0002] The development of the Internet has entered the information cloud era. Images, as an intuitive and convenient information carrier, are everywhere on the Internet. However, the security of multimedia image information is not guaranteed, and piracy, infringement, and malicious tampering are emerging in an endless stream. Based on this, a double watermark algorithm is proposed to determine the copyright owner of the digital image and authenticate the authenticity and integrity of the digital image.

[0003] The reversible data hiding algorithm based on dual images has attracted wide attention due to its high security and strong embedding ability. First, the binary secret data is converted into quinary and then embedded into a pixel pair. The experimental results show that the proposed method achieves the same embedding ability of the stego-image and slightly lower visual quality. However, the proposed method can completely restore the original cover image after the secret data is extracted. In order to further reduce the distortion of the stego-image, a novel dual-image reversible data hiding algorithm based on the center folding strategy is proposed. The center folding strategy (CFS) is used to re-encode the secret data to be embedded, and then the encoded data is embedded into two carrier images on average. The algorithm effectively reduces the distortion caused by embedding the secret data by folding the secret data. Under the same embedding capacity condition, the stego-image has better visual quality. The dual watermarking scheme for strengthening copyright protection embeds the visible watermark directly into the spatial domain of the host image and embeds the invisible watermark into the frequency domain of the host image. The dual robust watermarking algorithm embeds one watermark into the RGB color space of the host image and the other into the YCbCr color space of the host image. By embedding two watermarks in different color spaces, the algorithm's ability to resist multiple attacks is improved. These two algorithms have high robustness, which improves the algorithm's ability to protect image copyrights, but they cannot authenticate the integrity of the image. When the image is tampered with, the tampering cannot be detected.

[0004] Based on the above research, this paper proposes a dual watermarking algorithm in the wavelet transform domain based on improved image interpolation. The algorithm combines image interpolation and wavelet transform technology, expands the embedding capacity through image interpolation, and embeds the watermark in the low-frequency subband of the 3-level integer wavelet transform to protect the copyright. At the same time, the feature watermark is embedded in the middle and high frequency part of the 2-level transform to protect the image content, and the mean quantization method is used to successfully locate the tampering position. Summary of the invention

[0005] Purpose of the invention: In view of the problems existing in the prior art, the present invention provides a double watermarking method in the wavelet transform domain based on image interpolation, which improves the embedding capacity by using improved image interpolation, and uses the multi-layer decomposition and local characteristics of wavelet transform to skillfully embed watermarks in the low-frequency area of ​​the original image after the three-level transformation by changing the calculation method of wavelet coefficients to achieve the protection of image copyright. On this basis, the carrier image embedded with the watermark is further subjected to a three-level wavelet transform, and the coefficient correlation of the low-frequency part is extracted as an authentication watermark and embedded in the middle and high-frequency part of the two-level transformation to achieve the protection function of the image content and successfully locate the tampering position.

[0006] Technical solution: The present invention provides a double watermarking method in the wavelet transform domain based on image interpolation, which is characterized by comprising a watermark embedding method, specifically comprising the following steps:

[0007] Step 1: Perform Arnold transformation on the watermark R to obtain R′, and convert the transformed watermark R′ into a one-dimensional binary sequence R′ 1 ;

[0008] Step 2: Perform image interpolation on the original image O to obtain the interpolated image C;

[0009] Step 3: Decompose the interpolation image C by three-layer integer wavelet transform to obtain the third-level approximation sub-image LL 1 3 , approximate the third level subgraph LL 1 3 The coefficient is recorded as X, and X_avg is the LL adjacent to it. 1 3 The average value of the coefficients, if the coefficient X of the approximation sub-graph satisfies: X_avg≤X, then X=X_avg×(1+q), q represents the modified intensity value; if the coefficient X of the approximation sub-graph satisfies: X_avg>X, then X=X_avg×(1-q); the one-dimensional binary sequence R′ 1 and X according to the embedding formula P = X + R ′ 1 The new coefficient value P is used as a parameter to perform wavelet transform reconstruction to obtain an image embedded with a robust watermark; among them, the low-frequency subband LL 1 3 The superscript 1 represents the first three-layer wavelet transform decomposition;

[0010] Step 4: Perform three-layer integer wavelet transform decomposition on the image embedded with the robust watermark in step 3 again to obtain the approximate sub-image LL 2 3 and high frequency detail sub-image LH 2 2 , calculate the approximate subgraph LL 2 3The correlation between each coefficient and its neighboring coefficients in generates an authentication watermark; where the approximation subgraph LL 2 3 and high frequency detail sub-image LH 2 2 The superscript 2 represents the second three-layer wavelet transform decomposition;

[0011] Step 5: For the high-frequency detail sub-image LH 2 2 The coefficient of Divide the coefficients into groups of 4 and calculate the average value avg of each group. Then quantize the average value and embed the authentication watermark obtained in step 4 into it.

[0012] Step 6: After the authentication watermark is embedded, wavelet transform reconstruction is performed again to finally obtain the encrypted image W.

[0013] Furthermore, in step 1, the watermark R is subjected to Arnold transformation to obtain R′ as follows:

[0014]

[0015] In the formula, mod() is the modulus operation, (x, y) T is the watermark R pixel, (x', y') T is the transformed R′ pixel, and N is the order of the image matrix.

[0016] Furthermore, in step 2, image interpolation is performed on the original image I, and the specific process is as follows:

[0017] In the image interpolation process, the original image O of size w×h is interpolated and enlarged to generate an interpolated image C of size 2w×2h. The average value of the four closest reference pixels is used to replace the calculation of the diagonal interpolation pixels in the ENMI method, specifically:

[0018] Assume that 2×2 pixels in the original image O are used as reference pixels, and the interpolated image C after interpolation is divided into 3×3 overlapping blocks, and adjacent 3×3 overlapping blocks share one column or one row; the 2×2 reference pixels are used as the vertices of the 3×3 overlapping blocks, and among the pixels of the remaining 3×3 overlapping blocks, the average value of the four nearest reference pixels is used to calculate the interpolated pixels in the diagonal direction, and the interpolated pixels in the horizontal and vertical directions are given the average value of the two nearest reference pixels and the two adjacent diagonal interpolated pixels.

[0019] Furthermore, the specific interpolation calculation method is as follows:

[0020]

[0021] Among them, 1≤i<w / 2, 1≤j<h / 2, the symbol represents rounding down, O(i,j), O(i,j+1), O(i+1,j), O(i+1,j+1) represent 2×2 pixels in the original image O, i and j represent the horizontal and vertical coordinates of the pixels; C(2i-1,2j-1), C(2i,2j), C(2i-1,2j), C(2i,2j-1) represent the interpolated pixels of the interpolated image C.

[0022] Furthermore, the specific process of generating the authentication watermark in step 4 is as follows:

[0023] 1) Set They are the approximate part LL of the image embedded with the robust watermark after being decomposed by the wavelet layer transform. 2 1 , horizontal part HL 2 1 , vertical part LH 2 1 , diagonal part HH 2 1 The wavelet coefficients of each represent a pixel matrix image block of size 8×8 in the spatial domain before the three-layer wavelet decomposition of the image embedded with the robust watermark;

[0024] 2) Select the approximation subgraph LL 2 3 The correlation between the coefficients of and its adjacent coefficients is used as the authentication watermark to approximate the subgraph LL 2 3 The coefficients are scanned from left to right in sequence starting from the horizontal position. When one row is scanned, the direction is switched to the next row and scanned from right to left. This is repeated until all coefficients are scanned.

[0025] 3) Let the approximation subgraph LL 2 3 Two adjacent coefficients and The spacing between the two coefficients is S. If S<0, the correlation between the two coefficients is recorded as 1. If S≥0, the correlation between the two coefficients is recorded as 0, and the data set consisting of 0 and 1 is combined as the authentication watermark.

[0026] Furthermore, the number of average values ​​avg in step 5 is the same as the number of extracted authentication watermarks, both of which are Where 2w×2h is the size of the interpolated image C.

[0027] Furthermore, an image tampering detection method is also included, which is as follows:

[0028] Step 1: Downsample the encrypted image W to obtain the original encrypted image O′;

[0029] Step 2: Perform image interpolation on the original encrypted image O′ to obtain the interpolated image C′;

[0030] Step 3: Decompose the interpolation image C′ by three-layer integer wavelet transform to obtain the approximate sub-image LL 3 3 and high frequency detail sub-image LH 3 2 , where the approximation subgraph LL 3 3 and high frequency detail sub-image LH 3 2 The superscript 3 represents the third three-layer wavelet transform decomposition; the calculation approximation subgraph LL 3 3 The correlation between each coefficient and its neighboring coefficients generates an authentication watermark 1 for the high-frequency detail sub-image LH 3 2 Wavelet transform is performed to extract authentication watermark 2, and then authentication watermark 1 and authentication watermark 2 are compared to see if they are equal. If they are equal, it means that the image content has not been tampered with; if they are not equal, it means that the image has been tampered with.

[0031] Furthermore, the approximate subgraph LL is calculated 3 3 The correlation between each coefficient and its neighboring coefficients generates an authentication watermark, as follows:

[0032] 1) Set They are the approximate part LL of the interpolation image C′ after the three-layer integer wavelet transform decomposition. 3 1 , horizontal part HL 3 1 , vertical part LH 3 1 , diagonal part HH 3 1 The wavelet coefficients of , each represents a pixel matrix image block of size 8×8 in the spatial domain before the three-layer wavelet decomposition of the interpolated image C′;

[0033] 2) Select the approximation subgraph LL 3 3 The coefficient of The correlation with its adjacent coefficients is used as the authentication watermark for the approximation subgraph LL 3 3 The coefficients are scanned from left to right in sequence starting from the horizontal position. When one row is scanned, the direction is switched to the next row and scanned from right to left. This is repeated until all coefficients are scanned.

[0034] 3) Let the approximation subgraph LL 3 3Two adjacent coefficients and The spacing between the two coefficients is S′. If S′<0, the correlation between the two coefficients is recorded as 1. If S′≥0, the correlation between the two coefficients is recorded as 0. The data set consisting of 0 and 1 is combined as an authentication watermark.

[0035] Furthermore, for the high-frequency detail sub-image LH 3 2 The specific process of wavelet transform to extract authentication watermark 2 is as follows:

[0036] The LH after the third integer wavelet two-layer decomposition 3 2 The coefficient of The data are divided into groups of 4 and the average value avg′ of each group is calculated. The average value avg′ is then dequantized and analyzed to obtain the authentication watermark 2, thereby realizing the extraction of the authentication watermark.

[0037] Furthermore, a watermark extraction method is also included, which is as follows:

[0038] Step 1: Downsample the encrypted image W to obtain the original encrypted image O′;

[0039] Step 2: Perform image interpolation on the original encrypted image O′ to obtain the interpolated image C′;

[0040] Step 3: Decompose the interpolation image C′ by three-layer integer wavelet transform to obtain the approximate sub-image LL 3 3 , use X′ to represent the approximate subgraph LL 3 3 The coefficients of , where the approximation subgraph LL 3 3 The superscript 3 represents the third three-layer wavelet transform decomposition;

[0041] Step 4: X′_avg is the LL adjacent to its left and right 3 3 The average value of the coefficients, if the approximation subgraph LL 3 3 The coefficient X′ of the approximation sub-graph satisfies: X′_avg≤X′, then X′=X′_avg×(1+q), q represents the modified intensity value. If the coefficient X′ of the approximation sub-graph satisfies: X′_avg>X′, then X′=X′_avg×(1-q); according to LL 3 3 The coefficients X′ and X′_avg are extracted according to the formula R″ 1 =X′-X′_avg extracts the robust watermark coefficient R″ 1 ;

[0042] Step 5: Complete the extraction of the robust watermark and perform copyright authentication.

[0043] Beneficial effects:

[0044] The present invention combines the characteristics of image interpolation and wavelet transform. First, image interpolation is used to expand the embedding capacity. Then, the low-frequency subband LL3 coefficients in the 3-level integer wavelet transform are used to embed the watermark to protect the image copyright. Then, the correlation of the low-frequency subband coefficients, i.e., the characteristics, are used as watermarks to embed into the middle and high frequency parts of the 2-level transform to protect the image content. The mean quantization method is used to successfully locate the tampering position. Experiments have shown that the algorithm proposed in the present invention has high robustness and invisibility, successfully solves the problem of single function of most watermark algorithms, and strikes a balance between robustness and invisibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of watermark embedding of the present invention;

[0046] Figure 2 This is a flow chart of image tampering detection and watermark extraction of the present invention;

[0047] Figure 3 It is a framework diagram of the image interpolation method of the watermark embedding and extraction process of the present invention;

[0048] Figure 4 This is a schematic diagram of 3×3 overlapping blocks of an image according to the present invention;

[0049] Figure 5 This is an example diagram of the image interpolation process of the present invention;

[0050] Figure 6 It is a schematic diagram of integer wavelet transform decomposition of the present invention;

[0051] Figure 7 It is an experimental image of an embodiment of the present invention;

[0052] Figure 8 is the watermark image of the present invention;

[0053] Fig. 9 This is an image after a conventional attack in an embodiment of the present invention;

[0054] Fig.10 PSNR comparison of the conventional attack of the present invention;

[0055] Fig.11 An image of a geometric attack according to an embodiment of the present invention;

[0056] Fig.12 PSNR comparison of the geometric attack of the present invention;

[0057] Fig.13This is the initial tamper detection marking result diagram of the present invention. DETAILED DESCRIPTION

[0058] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0059] The invention discloses a double watermarking method in wavelet transform domain based on image interpolation, including a watermark embedding method, which specifically includes the following steps:

[0060] Step 1: Perform Arnold transformation on the watermark R to obtain R′, and convert the transformed watermark R′ into a one-dimensional binary sequence R′ 1 .

[0061]

[0062] In the formula, mod() is the modulus operation, (x, y) T is the watermark R pixel, (x', y') T is the transformed R′ pixel, and N is the order of the image matrix.

[0063] Step 2: Perform image interpolation on the original image O to obtain the interpolated image C.

[0064] Perform image interpolation on the original image I, see Figures 3 to 5 , the specific process is as follows:

[0065] In the image interpolation process, the original image O of size w×h is interpolated and enlarged to generate an interpolated image C of size 2w×2h. The average value of the four closest reference pixels is used to replace the calculation of the diagonal interpolation pixels in the ENMI method, specifically:

[0066] Set the 2×2 pixels in the original image O as reference pixels, see Figure 4 and Figure 5 , the interpolated image C is divided into 3×3 overlapping blocks, and adjacent 3×3 overlapping blocks share one column or one row. Figure 5 The red box in the figure is a 3×3 overlapping block, and the 2×2 reference pixels ( Figure 5 O(0,0), O(0,1), O(1,0), O(1,1)) in the 3×3 overlapping blocks are used as the vertices of the 3×3 overlapping blocks. Among the pixels of the remaining 3×3 overlapping blocks, the average value of the four nearest reference pixels is used to calculate the interpolated pixels in the diagonal direction, and the interpolated pixels in the horizontal and vertical directions are given the average value of the two nearest reference pixels and the two adjacent diagonal interpolated pixels.

[0067] The specific interpolation calculation method is as follows:

[0068]

[0069] Among them, 1≤i<w / 2, 1≤j<h / 2, the symbol represents rounding down, O(i,j), O(i,j+1), O(i+1,j), O(i+1,j+1) represent 2×2 pixels in the original image O, i and j represent the horizontal and vertical coordinates of the pixels; C(2i-1,2j-1), C(2i,2j), C(2i-1,2j), C(2i,2j-1) represent the interpolated pixels of the interpolated image C.

[0070] Specific examples in this implementation are as follows:

[0071] The reference pixel in the figure remains unchanged, such as Figure 4 The specific calculation method is C(2i-1,2j-1)=O(i,j). Then the average value of the four nearest reference pixels is used to calculate the interpolated pixel C(2i,2j) in the diagonal direction, as shown in Figure 4 The black blocks such as C(1,1) and C(1,3) are calculated in detail.

[0072]

[0073] Then, the interpolated pixels in the horizontal and vertical directions are given the average of the two nearest reference pixels and the two adjacent diagonal interpolated pixels, as Figure 4 The gray blocks such as C(0,1) and C(1,0) are calculated as follows:

[0074]

[0075] That is, C(0,0)=O(0,0), C(0,2)=O(0,1), C(2,0)=O(1,0), C(2,2)=O(1,1);

[0076] C(1,1)=(C(0,0)+C(0,2)+C(2,0)+C(2,2)) / 4, C(1,3)=(C(0,2)+C(0,4)+C(2,2)+C(2,4)) / 4, C(3,1)=(C(2,0)+C(2,2)+C(4,0)+C( 4,2)) / 4, C(3,3)=(C(2,2)+C(2,4)+C(4,2)+C(4,4)) / 4; C(0,1)=(C(0,0)+C(0,2)+C(1,1)) / 3, C(1,0)=(C(0,0)+C(1,1)+C(2,0)) / 3

[0077] C(1,2)=(C(0,2)+C(1,1)+C(1,3))+C(2,2)) / 4, C(2,1)=(C(1,1)+C(2,0)+C(3,1))+C(2,2)) / 4.

[0078] Step 3: Decompose the interpolation image C by three-layer integer wavelet transform to obtain the third-level approximation sub-image LL 1 3 , approximate the third level subgraph LL 1 3 The coefficient is recorded as X, and X_avg is the LL adjacent to it. 1 3 The average value of the coefficients, if the coefficient X of the approximation sub-graph satisfies: X_avg≤X, then X=X_avg×(1+q), q represents the modified intensity value; if the coefficient X of the approximation sub-graph satisfies: X_avg>X, then X=X_avg×(1-q); the one-dimensional binary sequence R′ 1 and X according to the embedding formula P = X + R ′ 1 The new coefficient value P is used as a parameter to perform wavelet transform reconstruction to obtain an image embedded with a robust watermark; among them, the low-frequency subband LL 1 3 The superscript 1 represents the first three-layer wavelet transform decomposition.

[0079] Step 4: Perform three-layer integer wavelet transform decomposition on the image embedded with the robust watermark in step 3 again to obtain the approximate sub-image LL 2 3 and high frequency detail sub-image LH 2 2 , calculate the approximate subgraph LL 2 3 The correlation between each coefficient and its neighboring coefficients in generates an authentication watermark; where the approximation subgraph LL 2 3 and high frequency detail sub-image LH 2 2 The superscript 2 represents the second three-layer wavelet transform decomposition.

[0080] The specific process of generating an authentication watermark is as follows:

[0081] 1) Set They are the approximate part LL of the image embedded with the robust watermark after being decomposed by the wavelet layer transform. 2 1 , horizontal part HL 2 1 , vertical part LH 2 1 , diagonal part HH 2 1The wavelet coefficients of each represent a pixel matrix image block of size 8×8 in the spatial domain before the three-layer wavelet decomposition of the image embedded with the robust watermark.

[0082] 2) Select the approximation subgraph LL 2 3 The correlation between the coefficients of and its adjacent coefficients is used as the authentication watermark to approximate the subgraph LL 2 3 The coefficients are scanned from left to right in sequence starting from the horizontal position. When one row is scanned, the direction is switched to the next row and scanned from right to left. This is repeated until all the coefficients are scanned.

[0083] 3) Let the approximation subgraph LL 2 3 Two adjacent coefficients and The spacing between the two coefficients is S. If S<0, the correlation between the two coefficients is recorded as 1. If S≥0, the correlation between the two coefficients is recorded as 0, and the data set consisting of 0 and 1 is combined as the authentication watermark.

[0084] Step 5: For the high-frequency detail sub-image LH 2 2 The coefficient of Divide the coefficients into groups of 4 and calculate the average value avg of each group. Then quantize the average value and embed the authentication watermark obtained in step 4 into it. The number of average values ​​avg is the same as the number of extracted authentication watermarks, both of which are Where 2w×2h is the size of the interpolated image C.

[0085] Step 6: After the authentication watermark is embedded, wavelet transform reconstruction is performed again to finally obtain the encrypted image W.

[0086] Since both the robust watermark and the authentication watermark of the present invention utilize wavelet coefficients, the authentication watermark should be extracted first and then the robust watermark. The image tampering detection method is also included, which is as follows:

[0087] Step 1: Downsample the encrypted image W to obtain the original encrypted image O′.

[0088] Step 2: Perform image interpolation on the original encrypted image O′ to obtain the interpolated image C′.

[0089] Step 3: Decompose the interpolation image C′ by three-layer integer wavelet transform to obtain the approximate sub-image LL 3 3 and high frequency detail sub-image LH 3 2 , where the approximation subgraph LL 3 3and high frequency detail sub-image LH 3 2 The superscript 3 represents the third three-layer wavelet transform decomposition; the calculation approximation subgraph LL 3 3 The correlation between each coefficient and its neighboring coefficients generates an authentication watermark 1 for the high-frequency detail sub-image LH 3 2 Wavelet transform is performed to extract authentication watermark 2, and then authentication watermark 1 and authentication watermark 2 are compared to see if they are equal. If they are equal, it means that the image content has not been tampered with; if they are not equal, it means that the image has been tampered with.

[0090] Compute the approximate subgraph LL 3 3 The correlation between each coefficient and its neighboring coefficients generates an authentication watermark, as follows:

[0091] 1) Set They are the approximate part LL of the interpolation image C′ after the three-layer integer wavelet transform decomposition. 3 1 , horizontal part HL 3 1 , vertical part LH 3 1 , diagonal part HH 3 1 The wavelet coefficients of , each represent a pixel matrix image block of size 8×8 in the spatial domain before the three-layer wavelet decomposition of the interpolated image C′.

[0092] 2) Select the approximation subgraph LL 3 3 The coefficient of The correlation with its adjacent coefficients is used as the authentication watermark for the approximation subgraph LL 3 3 The coefficients are scanned from left to right in sequence starting from the horizontal position. When one row is scanned, the direction is switched to the next row and scanned from right to left. This is repeated until all the coefficients are scanned.

[0093] 3) Let the approximation subgraph LL 3 3 Two adjacent coefficients and The spacing between the two coefficients is S′. If S′<0, the correlation between the two coefficients is recorded as 1. If S′≥0, the correlation between the two coefficients is recorded as 0. The data set consisting of 0 and 1 is combined as an authentication watermark.

[0094] For the high frequency detail sub-image LH 3 2 The specific process of wavelet transform to extract authentication watermark 2 is as follows:

[0095] The LH after the third integer wavelet two-layer decomposition 3 2 The coefficient of The data are divided into groups of 4 and the average value avg′ of each group is calculated. The average value avg′ is then dequantized and analyzed to obtain the authentication watermark 2, thereby realizing the extraction of the authentication watermark.

[0096] The watermark extraction method is as follows:

[0097] Step 1: Downsample the encrypted image W to obtain the original encrypted image O′.

[0098] Step 2: Perform image interpolation on the original encrypted image O′ to obtain the interpolated image C′.

[0099] Step 3: Decompose the interpolation image C′ by three-layer integer wavelet transform to obtain the approximate sub-image LL 3 3 , use X′ to represent the approximate subgraph LL 3 3 The coefficients of , where the approximation subgraph LL 3 3 The superscript 3 represents the third three-layer wavelet transform decomposition.

[0100] Step 4: X′_avg is the LL adjacent to its left and right 3 3 The average value of the coefficients, if the approximation subgraph LL 3 3 The coefficient X′ of the approximation sub-graph satisfies: X′_avg≤X′, then X′=X′_avg×(1+q), q represents the modified intensity value. If the coefficient X′ of the approximation sub-graph satisfies: X′_avg>X′, then X′=X′_avg×(1-q); according to LL 3 3 The coefficients X′ and X′_avg are extracted according to the formula R″ 1 =X′-X′_avg extracts the robust watermark coefficient R″ 1 .

[0101] Step 5: Complete the extraction of the robust watermark and perform copyright authentication.

[0102] In order to verify the superiority of the proposed algorithm, four standard grayscale images of House, Lack, Pepper, and Plane with a size of 512×512 are selected as test images. Figure 7As shown. This simulation experiment is based on Windows 11 system and is carried out using Matlab 2016a. PSNR, NC value and correct detection rate are used as performance evaluation indicators. The algorithm in this paper is compared with the algorithms in references

[17] ,

[18] , and

[19] to fully and objectively reflect the visual quality and detection positioning of this algorithm.

[0103] References

[17] Chen Xing'e. Research on dual digital watermarking technology for image copyright protection and content authentication [J]. Journal of Langfang Normal University (Natural Science Edition), 2013, 13(4): 38-43.

[0104] Literature

[18] Su GD,Chang CC,Chen C CA hybrid-Sudoku based fragilewatermarking scheme for image tampering detection[J].Multimedia Tools andApplications,2021,80:12881-12903.

[0105] Literature

[19] Sahu A KA logistic map basedblind and fragile watermarking for tamper detection and localization in images[J]. Journal of AmbientIntelligence and Humanized Computing, 2022,13(8):3869-3881.

[0106] 1 Watermark Robustness

[0107] The robustness of this experiment is tested by using the algorithm in the literature

[17] to perform conventional attacks and geometric attacks on four standard grayscale watermarked images, and the PSNR value and NC value under each attack are calculated.

[0108] 1.1 Conventional Attacks

[0109] The image is subjected to conventional attacks such as Gaussian noise (0, 0.01), Gaussian noise (0.01, 0.1), JPEG compression quality factor 5, JPEG compression quality factor 10, 3×3 median filtering, and 5×5 median filtering. The following are the specific operation steps.

[0110] Gaussian noise with a mean of 0 and a variance of 0.01 and Gaussian noise with a mean of 0.01 and a variance of 0.1 are added to the watermarked image. From the results, it can be seen that although the addition of Gaussian noise causes a significant decrease in image quality, it does not affect the extraction of the watermark. The watermarked image can still be extracted and recognized, and the similarity of the extracted watermark can reach 0.8453 and 0.7892 respectively. It can be seen that this algorithm has a certain robustness in image noise processing. The results are shown in Fig. 9 (a) shows. The watermark image is compressed by JPEG, and the compression quality factors are 5 and 10 respectively. The watermark is extracted from the compressed watermark image, and the similarity of the extracted watermark reaches 0.8453 and 0.8341 respectively. The JPEG compression attack test results show that the algorithm has good robustness to high-quality compression. The results are shown in Fig. 9 (b) As shown. The images are filtered with 3×3 and 5×5 median filters respectively, and the similarities of the extracted watermarks can reach 0.8962 and 0.8672 respectively. Fig. 9 (c) as shown.

[0111] Table 1 Comparison of watermark robustness test results under conventional attacks

[0112]

[0113] Reference

[17] is a robust digital watermark based on singular value decomposition for image copyright protection. This paper embeds the robust watermark through three-layer wavelet transform and then conducts conventional attacks. Fig.10 It can be seen that the PSNR and NC values ​​of the proposed method are higher than those of the algorithm proposed in the literature

[17] , whether after Gaussian noise, JPEG compression or median filter attack. Since this paper adopts the two-dimensional wavelet transform method to embed the watermark into the medium and high frequency subbands of the image, the medium and high frequency subbands are relatively less affected by image compression, noise and other transformations. Therefore, the watermark can be successfully embedded while maintaining high image quality. In order to ensure the accuracy and diversity of the experiment, this paper also changes the parameters of the same attack to form a comparative test of different parameters and different attacks, and obtains the data shown in Table 1. The experiment shows that the image quality of the proposed algorithm after conventional attacks is better than that of the literature

[17] .

[0114] 1.2 Geometry Attack

[0115] Geometric attack refers to modifying the position of an image through transformation, so that the original image and the transformed image lose synchronization in spatial position. This paper uses image shearing, image 30-degree rotation, and image 50-degree rotation as geometric attacks to evaluate the robustness of the watermarking algorithm.

[0116] The watermarked image was cropped and the upper left corner of the watermarked image was cropped. This caused considerable damage to the original image, but the watermark was still detected and the similarity reached 0.7547. The cropped watermarked image and the extracted watermark are shown in Fig.11 (a) shows that the original image is rotated by 30 degrees and 50 degrees respectively, and the correlation NC is 0.7968 and 0.7848 respectively. The rotated watermarked image and the extracted watermark are shown in Fig.11 (b) as shown.

[0117] Table 2 Comparison of geometric attack watermark robustness test results

[0118]

[0119]

[0120] This paper embeds a robust watermark through three-layer wavelet transform and then conducts geometric attack. As can be seen from Table 2, the PSNR and NC values ​​of this method are higher than those of the algorithm proposed in the literature

[17] after image shearing or image rotation attack. This is because the algorithm in this paper embeds the watermark through wavelet coefficients, and the impact on the high-frequency sub-bands of the image after image shearing or image rotation is relatively small. Therefore, the PSNR and NC values ​​are relatively high, and the image quality is relatively high.

[0121] 2Tamper Detection

[0122] The present invention performs simple attacks such as collage attack, cutting attack and text addition on the watermarked image, and calculates its false alarm rate (Rfa), missed alarm rate (Rma) and false detection rate (Rfd). A copy of the image block of size 100×100 belonging to the experimental image is cut and placed on a certain part of the experimental image, and then the tampering location is detected, such as Fig.13 (a) After the experimental image is subjected to a shear attack with a size of 150×150 image blocks, the tampering positioning is detected. The initial detection positioning map is shown in Fig.13 (b) As shown. After adding text to the upper left corner of the experimental image, the tampering positioning is detected. The initial detection positioning map is shown in Fig.13 (c) as shown.

[0123] As can be seen from Table 3, the false alarm rate and false detection rate of the reference

[18] are slightly better than those of the method selected in this paper. This is because the algorithm uses a two-layer detection scheme to greatly reduce the false alarm rate. The reference

[19] uses the principle of Logistic mapping to construct or embed fragile watermarks, just like the method selected in this paper. Therefore, the performance of the two methods is very similar. From the simulation results in Table 3, it can be seen that the tampering location capability of the algorithm proposed in this paper is also very outstanding. Since the watermark is embedded in the mid-high frequency sub-band of the image, it has strong robustness to common attacks such as image compression, noise, and filtering, and can effectively detect image tampering. The dual digital watermarking algorithm based on the wavelet transform domain not only increases the anti-attack capability of the watermark, but also makes the watermark have better robustness and invisibility. More importantly, on this basis, a semi-fragile watermark that can detect the tampering position of the image content and realize accurate positioning is embedded, so that the carrier image containing the robust watermark and the semi-fragile watermark is protected by copyright authentication and content authenticity.

[0124] Table 3 Comparison of tamper detection performance

[0125]

[0126] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A double watermarking method in wavelet transform domain based on image interpolation, characterized in that: The invention comprises a watermark embedding method, which specifically comprises the following steps: Step 1: Perform Arnold transformation on the watermark R to obtain R′, and convert the transformed watermark R′ into a one-dimensional binary sequence R′1; Step 2: Perform image interpolation on the original image O to obtain the interpolated image C; Step 3: Decompose the interpolation image C by three-layer integer wavelet transform to obtain the third-level approximation sub-image LL 1 3. Approximate the third level subgraph LL 1 The coefficient of 3 is recorded as X, and X_avg is the LL adjacent to it. 1 3, if the coefficient X of the approximation sub-image satisfies: X_avg≤X, then X=X_avg×(1+q), q represents the modified intensity value; if the coefficient X of the approximation sub-image satisfies: X_avg>X, then X=X_avg×(1-q); the one-dimensional binary sequence R′1 and X are calculated according to the embedding formula P=X+R′1, and the new coefficient value P is used as a parameter to perform wavelet transform reconstruction to obtain an image embedded with a robust watermark; among them, the low-frequency sub-band LL 1 The superscript 1 in 3 represents the first three-layer wavelet transform decomposition; Step 4: Perform three-layer integer wavelet transform decomposition on the image embedded with the robust watermark in step 3 again to obtain the approximate sub-image LL 2 3 and high frequency detail sub-image LH 2 2. Calculate the approximate subgraph LL 2 The correlation between each coefficient and its neighboring coefficients in 3 generates an authentication watermark; among them, the approximation subgraph LL 2 3 and high frequency detail sub-image LH 2 The superscript 2 in 2 represents the second three-layer wavelet transform decomposition; Step 5: For the high-frequency detail sub-image LH 2 The coefficient I2 of 2 2 Divide the coefficients into groups of 4 and calculate the average value avg of each group. Then quantize the average value and embed the authentication watermark obtained in step 4 into it. Step 6: After the authentication watermark is embedded, wavelet transform reconstruction is performed again to finally obtain the encrypted image W.

2. According to claim 1, a double watermarking method based on wavelet transform domain of image interpolation is characterized in that: In step 1, the watermark R is transformed by Arnold to obtain R′ as follows: In the formula, mod() is the modulus operation, (x, y) T is the watermark R pixel, (x', y') T is the transformed R′ pixel, and N is the order of the image matrix.

3. The double watermarking method based on wavelet transform domain of image interpolation according to claim 1 is characterized in that: In step 2, image interpolation is performed on the original image I, and the specific process is as follows: In the image interpolation process, the original image O of size w×h is interpolated and enlarged to generate an interpolated image C of size 2w×2h. The average value of the four closest reference pixels is used to replace the calculation of the diagonal interpolation pixels in the ENMI method, specifically: Assume that 2×2 pixels in the original image O are used as reference pixels, and the interpolated image C after interpolation is divided into 3×3 overlapping blocks, and adjacent 3×3 overlapping blocks share one column or one row; the 2×2 reference pixels are used as the vertices of the 3×3 overlapping blocks, and among the pixels of the remaining 3×3 overlapping blocks, the average value of the four nearest reference pixels is used to calculate the interpolated pixels in the diagonal direction, and the interpolated pixels in the horizontal and vertical directions are given the average value of the two nearest reference pixels and the two adjacent diagonal interpolated pixels.

4. The double watermarking method based on wavelet transform domain of image interpolation according to claim 3 is characterized in that: The specific interpolation calculation method is as follows: Among them, 1≤i<w / 2, 1≤j<h / 2, the symbol represents rounding down, O(i,j), O(i,j+1), O(i+1,j), O(i+1,j+1) represent 2×2 pixels in the original image O, i and j represent the horizontal and vertical coordinates of the pixels; C(2i-1,2j-1), C(2i,2j), C(2i-1,2j), C(2i,2j-1) represent the interpolated pixels of the interpolated image C.

5. The double watermarking method based on wavelet transform domain of image interpolation according to claim 1 is characterized in that: The specific process of generating the authentication watermark in step 4 is as follows: 1) Set They are the approximate part LL of the image embedded with the robust watermark after being decomposed by the wavelet layer transform. 2 1. Horizontal part HL 2 1. Vertical part LH 2 1. Diagonal part HH 2 1, each represents an 8×8 pixel matrix image block in the spatial domain before the three-layer wavelet decomposition of the image embedded with the robust watermark; 2) Select the approximation subgraph LL 2 The correlation between the coefficient of 3 and its adjacent coefficients is used as the authentication watermark for the approximate subgraph LL 2 The coefficients of 3 are scanned from left to right in sequence starting from the horizontal position. When one row is scanned, the direction is switched to the next row and scanned from right to left. This is repeated until all coefficients are scanned. 3) Let the approximation subgraph LL 2 3Two adjacent coefficients and The spacing between the two coefficients is S. If S<0, the correlation between the two coefficients is recorded as 1. If S≥0, the correlation between the two coefficients is recorded as 0, and the data set consisting of 0 and 1 is combined as the authentication watermark.

6. The double watermarking method based on wavelet transform domain of image interpolation according to claim 1 is characterized in that: The number of average values ​​avg in step 5 is the same as the number of extracted authentication watermarks, both of which are Where 2w×2h is the size of the interpolated image C.

7. A double watermarking method in wavelet transform domain based on image interpolation according to any one of claims 1 to 6, characterized in that: It also includes image tampering detection methods, as follows: Step 1: Downsample the encrypted image W to obtain the original encrypted image O′; Step 2: Perform image interpolation on the original encrypted image O′ to obtain the interpolated image C′; Step 3: Decompose the interpolation image C′ by three-layer integer wavelet transform to obtain the approximate sub-image LL 3 3 and high frequency detail sub-image LH 3 2, where the approximation subgraph LL 3 3 and high frequency detail sub-image LH 3 The superscript 3 in 2 represents the third three-layer wavelet transform decomposition; the approximate subgraph LL is calculated 3 The correlation between each coefficient and its neighboring coefficients in 3 generates an authentication watermark 1 for the high-frequency detail sub-image LH 3 2 Perform wavelet transform to extract authentication watermark 2, and then compare whether authentication watermark 1 and authentication watermark 2 are equal. If they are equal, it means that the image content has not been tampered with; if they are not equal, it means that the image has been tampered with.

8. The wavelet transform domain dual watermarking method based on image interpolation according to claim 7 is characterized in that: Compute the approximate subgraph LL 3 The correlation between each coefficient in 3 and its neighboring coefficients generates an authentication watermark, as follows: 1) Set They are the approximate part LL of the interpolation image C′ after the three-layer integer wavelet transform decomposition. 3 1. Horizontal part HL 3 1. Vertical part LH 3 1. Diagonal part HH 3 1, each represents a pixel matrix image block of size 8×8 in the spatial domain before the three-layer wavelet decomposition of the interpolated image C′; 2) Select the approximation subgraph LL 3 Coefficient of 3 The correlation with its adjacent coefficients is used as the authentication watermark for the approximation subgraph LL 3 The coefficients of 3 are scanned from left to right in sequence starting from the horizontal position. When one row is scanned, the direction is switched to the next row and scanned from right to left. This is repeated until all coefficients are scanned. 3) Let the approximation subgraph LL 3 3Two adjacent coefficients and The spacing between the two coefficients is S′. If S′<0, the correlation between the two coefficients is recorded as 1. If S′≥0, the correlation between the two coefficients is recorded as 0. The data set consisting of 0 and 1 is combined as an authentication watermark.

9. The double watermarking method based on wavelet transform domain of image interpolation according to claim 7 is characterized in that: For the high frequency detail sub-image LH 3 2 The specific process of wavelet transform to extract authentication watermark 2 is as follows: The LH after the third integer wavelet two-layer decomposition 3 The coefficient of 2 The data are divided into groups of 4 and the average value avg′ of each group is calculated. The average value avg′ is then dequantized and analyzed to obtain the authentication watermark 2, thereby realizing the extraction of the authentication watermark.

10. A double watermarking method in wavelet transform domain based on image interpolation according to any one of claims 1 to 6, characterized in that: It also includes a watermark extraction method, as follows: Step 1: Downsample the encrypted image W to obtain the original encrypted image O′; Step 2: Perform image interpolation on the original encrypted image O′ to obtain the interpolated image C′; Step 3: Decompose the interpolation image C′ by three-layer integer wavelet transform to obtain the approximate sub-image LL 3 3. Use X′ to represent the approximation subgraph LL 3 3 coefficients, where the approximation subgraph LL 3 The superscript 3 in 3 represents the third three-layer wavelet transform decomposition; Step 4: X′_avg is the LL adjacent to its left and right 3 3 coefficient average, if the approximation subgraph LL 3 3, if the coefficient X′ satisfies: X′_avg≤X′, then X′=X′_avg×(1+q), where q represents the modified intensity value. If the coefficient X′ of the approximation subgraph satisfies: X′_avg>X′, then X′=X′_avg×(1-q); according to LL 3 The coefficient X′ and X′_avg of 3 are used to extract the robust watermark coefficient R″1 according to the extraction formula R″1=X′-X′_avg; Step 5: Complete the extraction of the robust watermark and perform copyright authentication.

Images