Screen-shooting-resistant image right identification embedding method, extraction method and system
By embedding the right confirmation mark of anti-screen shooting in the image, using SIFT intensity and texture value screening, sharpening enhancement and DCT transformation, combined with BCH encoding, the problem that traditional watermarks are difficult to resist screen shooting attacks is solved, and efficient copyright protection is achieved and extraction accuracy.
Patent Information
- Application Number
- CN202410995137.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-15
AI Technical Summary
Traditional robust image watermarking schemes are difficult to resist screen shooting attacks, making it difficult to achieve digital image copyright protection.
The image confirmation mark embedding method that is anti-screen shooting is adopted, and the SIFT intensity calculation, texture value screening, sharpening enhancement and weighting mixing is combined with DCT transformation and BCH encoding to ensure that the confirmation mark can still be correctly extracted after screen shooting.
Improves the robustness and extraction accuracy of image copyright protection, and can effectively protect the copyright of digital images under screen shooting attacks.
Smart Images

Figure CN118967422A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of digital watermarking, and in particular to an image right confirmation mark embedding method, an extraction method and a system for resisting screen shooting. Background Art
[0002] With the continuous development of multimedia technology and the Internet, the widespread dissemination and circulation of digital products such as audio and video works, digital images, and electronic publications in electronic devices has become more convenient than ever before. Although this trend has greatly enriched people's information acquisition channels and entertainment life, it has also greatly reduced the cost of copyright infringement, leading to the intensification of copyright protection issues. In this context, academia, industry, and the international level are increasingly paying attention to how to effectively protect copyrights and ensure the secure transmission of information in a network environment. How to deal with the problems of infringement, piracy, and arbitrary tampering of digital content during network dissemination has become an important research focus.
[0003] In order to deal with the problem of copyright protection of digital works, encryption technology has become one of the common means. However, traditional encryption methods face many challenges. The first problem is that once it is clearly stated that the data is encrypted, it means that it contains sensitive information, which attracts the attention of attackers. With the rapid advancement of computing power today, advanced cracking technology has made it increasingly difficult to ensure the security of data with encryption keys alone. Once the key is indexed and decrypted, the digital content will lose its effective protection barrier and face the risk of being arbitrarily disseminated and tampered with. At the same time, in today's network communication process, the damage and disturbance that may be suffered are more complex, and a more robust and covert way is needed to protect the copyright of digital information.
[0004] In the face of the lossy channel transmission mentioned above, image watermarking technology is a more effective means. As a branch of steganography, digital watermarking technology embeds copyright information into the image carrier covertly, making it difficult for the naked eye to detect the difference before and after embedding. Compared with traditional steganography, digital watermarking technology focuses more on embedding information in the image to achieve a more sophisticated copyright protection goal. This technology embeds copyright information into the image carrier as a digital watermark covertly, ensuring that it is difficult for the naked eye to observe obvious changes before and after embedding, and at the same time ensuring that the watermark information can still be fully extracted after being attacked. When the protected image faces copyright disputes, the copyright authentication is performed by extracting the watermark information therein, thereby achieving effective copyright protection. This technology has a certain degree of anti-attack ability and provides a feasible solution for the long-term protection of digital works. After the sender embeds the watermark information into the carrier image, it is transmitted through the lossy channel and the watermark is extracted at the receiver to finally obtain the effective watermark information. This process needs to resist various types of image processing attacks and geometric perturbations, which are impossible for image encryption methods.
[0005] Take an art exhibition as an example. All works of art on display are protected by copyright, and visitors should not take photos and disseminate them at will. However, after the exhibition, if a photo that is extremely similar to a work is found on the Internet, but it is not certain whether it is a stolen photo of the exhibition, the image watermark technology that resists screen shooting can quickly confirm the copyright information in the photo, thereby effectively protecting the copyright of the work. In-depth research in this field will promote the continuous development of digital watermark technology and better respond to the increasingly complex copyright protection needs.
[0006] Some scholars in my country have proposed a robust watermarking scheme (SSR) for screen shooting. This scheme designs an intensity-based scale-invariant feature transform algorithm (I-SIFT), which is specifically used to accurately identify the embedded position of the watermark in the distorted image; at the same time, a small-size template algorithm is used to repeatedly embed the watermark information in different areas to ensure the watermark survival rate; in order to further ensure the extraction accuracy, an extraction algorithm based on cross-validation is also used.
[0007] Most traditional robust image watermarking schemes only target possible attacks on the image itself, such as cropping, scaling, noise, JPEG compression, etc. However, with the rapid development of mobile devices and camera technology, people record digital information on electronic screens by shooting and posting works on the Internet, which indirectly leads to the possibility of copyright infringement. In this case, the attack on the image is also called screen shooting attack. After shooting, digital images are usually damaged by both electronic and screen shooting channels. Traditional watermarking schemes are almost unable to resist such attacks, which is a major challenge in the field of digital image watermarking.
[0008] In view of this, an image authentication identification embedding method, extraction method and system resistant to screen shooting are proposed. Summary of the invention
[0009] The purpose of the present invention is to provide an image authentication mark embedding method, extraction method and system that are resistant to screen shooting, which can embed an authentication mark in an image, and the authentication mark can still be correctly extracted after being photographed by an electronic device, thereby ensuring that the image copyright can be effectively protected in the scene of voyeurism.
[0010] The above technical objectives of the present invention are achieved through the following technical solutions:
[0011] A method for embedding an image authentication mark that resists screen shooting comprises the following steps:
[0012] Strength calculation: extract key points, use the absolute value of the Gaussian difference of the key points as the SIFT strength of the point, calculate the strength of each key point, sort them in descending order, and select the first n feature points as candidate feature points;
[0013] Texture value calculation: Calculate the texture value of the 64×64 area around the candidate feature points, sort them in descending order, and use the preset number of feature points with the highest ranking as the final embedding center;
[0014] Sharpening: sharpen the surrounding area of the feature point determined as the embedding center;
[0015] Weighted blending: After sharpening, choose to weight the original area with the sharpened area in a specific ratio;
[0016] The ownership confirmation logo is embedded.
[0017] In a preferred embodiment, the embedding process of the ownership confirmation mark includes:
[0018] Reconstruction of the title confirmation sequence: Reconstruct the 64-bit 0 / 1 title confirmation sequence into an 8×8 title confirmation matrix;
[0019] Split the embedding region and perform DCT transform: Select a 64×64 embedding region in the image and split it into 64 8×8 sub-blocks;
[0020] Select DCT coefficients: Apply discrete cosine transform to each 8×8 sub-block and select the intermediate frequency coefficients C1 and C2 at the two positions (4,5) and (5,4) in each DCT block;
[0021] Modify DCT coefficients;
[0022] Embedding: Perform inverse DCT transform on each DCT block of the image to complete the identification embedding process;
[0023] Repeat the embedding process: Repeat the above steps for the remaining 64×64 areas to be embedded until all the embedding areas have completed the embedding of the ownership confirmation logo.
[0024] In a preferred embodiment, a key point is defined as p, and the intensity In(p) of the key point p is specifically defined as:
[0025] In(p)=|D(x,y,σ)|
[0026] Among them, D(x,y,σ) represents the Gaussian difference value of the key point, (x,y) is the coordinate in the image, and σ represents the variance of the Gaussian filter kernel.
[0027] In a preferred embodiment, the sharpening process is performed by convolving the image with a sharpening kernel, and the sharpening kernel is as follows:
[0028]
[0029] Among them, α is a parameter that controls the sharpening strength.
[0030] An image ownership identification embedding system resistant to screen shooting, comprising:
[0031] Strength calculation unit: used to extract key points, take the absolute value of Gaussian difference of key points as the SIFT strength of the point, calculate the strength of each key point, sort them in descending order, and select the first n feature points as candidate feature points;
[0032] Texture value calculation unit: used to calculate the texture value of the 64×64 area around the candidate feature points, sort them in descending order, and use the preset number of feature points with the highest ranking as the final embedding center;
[0033] Sharpening unit: used to sharpen the surrounding area of the feature point determined as the embedding center;
[0034] Weighted blending unit: used to select the original area and the sharpened area for weighted blending in a specific ratio after sharpening;
[0035] The title confirmation logo embedding unit is used to embed the title confirmation logo.
[0036] A method for extracting an image ownership identification that is resistant to screen shooting comprises the steps of:
[0037] Image acquisition: obtaining a title confirmation image to be extracted, wherein the title confirmation image to be extracted is obtained by reshooting the title confirmation image obtained by using the title confirmation image embedding method against screen shooting.
[0038] Image vertex confirmation: Receive the four blurred vertices of the target image in the screen shot image, use them as the center of the circle, and perform the Shi-Tomasi corner detection algorithm in a circular area with a radius of 10 pixels to find the accurate positions of the four vertices of the image;
[0039] Image correction: After obtaining the coordinates of the four vertices of the image containing the title confirmation mark in the photo, perform perspective distortion correction on it, and based on the result, correct the image in the photo to obtain a complete image containing the title confirmation mark;
[0040] Key point confirmation: select feature points within a preset range of intensity ranking in the image, calculate the regional texture values of these feature points, and take out feature points within a preset range of texture ranking as candidate key points;
[0041] Confirmation of the property rights identification area: Traverse the 3×3 area around each candidate key point to obtain the area that may contain the property rights identification, which is defined as the area to be extracted;
[0042] Extract the property rights identification.
[0043] In a preferred embodiment, the process of extracting the ownership confirmation mark includes:
[0044] Divide the area to be extracted into blocks, and divide the area of the title confirmation mark to be extracted into 8×8 sub-blocks, and the size of each sub-block is 8×8 pixels;
[0045] Perform DCT transform on the sub-blocks: Perform discrete cosine transform on each 8×8 sub-block. In the DCT coefficient matrix of these sub-blocks, select the intermediate frequency coefficients at positions (4,5) and (5,4), which are denoted as C1 and C2 respectively;
[0046] Extraction: After obtaining the intermediate frequency coefficients C1 and C2 of each sub-block, these coefficients are processed to extract an 8×8 identity matrix;
[0047] Extract and flatten the property rights identification matrix of all regions: Repeat the above steps for all block regions to extract the 8×8 property rights identification matrix of each region. Then, flatten these 8×8 property rights identification matrices in one dimension to obtain a linear property rights identification sequence.
[0048] Screening the most similar pairs of title confirmation identifiers: Compare the title confirmation identifier sequences between different groups in pairs, and screen out the most similar pair of title confirmation identifier sequences as candidate title confirmation identifier pairs;
[0049] Perform BCH decoding to obtain the final authentication mark: select any authentication mark sequence from the candidate authentication mark pair, perform BCH decoding operation, thereby restoring the original authentication mark information and extracting the authentication mark hidden in the image.
[0050] In a preferred embodiment, when screening the most similar authentication identification pairs, the comparison method is to calculate the cosine similarity between the two authentication identification sequences, and finally select the most similar pair of authentication identification sequences as a candidate pair to continue to execute subsequent steps.
[0051] In a preferred embodiment, the specific process of image correction includes: setting the coordinates of the four vertices before correction to P1 (x1, y1), P2 (x2, y2), P3 (x3, y3) and P4 (x4, y4), and the coordinates of the four vertices after correction to P'1 (x'1, y'1), P'2 (x'2, y'2), P'3 (x'3, y'3) and P'4 (x'4, y'4), respectively, by substituting them into the following formulas:
[0052]
[0053] After obtaining the eight equations, the eight variables a0, a1, a2, b0, b1, b2, c1, and c2 are solved jointly, from which the mapping formula for perspective distortion is obtained. The image in the photo is corrected using this formula, and finally a complete image containing the property rights identification is obtained.
[0054] An image right confirmation mark extraction system resistant to screen shooting, comprising
[0055] Image acquisition unit: used for acquiring the to-be-extracted ownership confirmation identification picture, wherein the to-be-extracted ownership confirmation identification picture is obtained by reshooting the ownership confirmation identification picture obtained by using the anti-screen-shooting image ownership confirmation identification embedding method as described in any one of claims 1-4.
[0056] Image vertex confirmation unit: used to receive the four fuzzy vertices of the target image in the screen shot image, and use them as the center of the circle to perform the Shi-Tomasi corner detection algorithm in a circular area with a radius of 10 pixels to find the accurate positions of the four vertices of the image;
[0057] Image correction unit: used to perform perspective distortion correction on the image containing the title confirmation mark in the photo after obtaining the coordinates of the four vertices of the image, and correct the image in the photo according to the result to obtain a complete image containing the title confirmation mark;
[0058] Key point confirmation unit: used to select feature points within a preset range of intensity ranking in the image, calculate the regional texture values of these feature points, and take out feature points within the preset range of texture ranking as candidate key points;
[0059] Title confirmation area confirmation unit: used to traverse the 3×3 area around each candidate key point to obtain the area that may contain the title confirmation mark, which is defined as the area to be extracted;
[0060] The title confirmation identification extraction unit is used to extract the title confirmation identification.
[0061] Compared with the prior art, the present invention has the following advantages:
[0062] (1) This scheme first performs BCH encoding on the copyright sequence, then uses I-SIFT and regional texture values to filter out the embedded regions of the copyright confirmation logo in the image, and performs sharpening enhancement, followed by embedding of local repeated copyright confirmation logos. In the extraction stage, the four blurred vertices of the image are first manually selected, and then the most likely corner points nearby are found through the corner detection algorithm, and perspective transformation is performed based on these corner points to obtain the corrected screen shot image; then the corrected image is located containing the copyright confirmation logo area. Unlike the embedding process, the extraction stage will select multiple additional regions to improve the extraction accuracy, extract the copyright confirmation logo from these regions, and select the most similar copyright confirmation logo as the final copyright confirmation logo through cross-validation. The final copyright confirmation logo is decoded through BCH to obtain the copyright number.
[0063] (2) A method for selecting embedding regions that combines I-SIFT and texture complexity is proposed. This method first uses the I-SIFT algorithm to select feature points with higher intensity, and then calculates the texture complexity of the area around these points for secondary screening, thereby selecting areas that are more suitable for embedding the property rights identification.
[0064] (3) A sharpening and enhancement method for embedded regions is proposed. This method sharpens and enhances the feature region and performs weighted fusion with the original region. Experiments have shown that this method improves the positioning accuracy of the region in the subsequent property rights identification extraction process.
[0065] (4) A screen shot correction method combined with a corner detection algorithm is proposed. The corner detection algorithm is automatically used to perform a range search around manually selected image vertices to find more accurate image vertices. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] Figure 1 Schematic diagram of a smooth texture area (left) and a rough texture area (right).
[0067] Figure 2 Schematic diagram of I-SIFT-based filtering (left) and filtering combined with regional texture values (right).
[0068] Figure 3 It is a schematic diagram of the process of embedding the property rights identification.
[0069] Figure 4 It is a schematic diagram of the screen capture correction process.
[0070] Figure 5 It is a schematic diagram of the corner detection process.
[0071] Figure 6 It is a schematic diagram of the process of extracting the property rights identification.
[0072] Figure 7 This is an example of the shooting environment and equipment placement for the experiment.
[0073] Figure 8 It is a schematic diagram of SIFT intensity values before and after sharpening enhancement.
[0074] Fig. 9 It is a schematic diagram of the PSNR value of the sharpened enhanced image and the original image.
[0075] Fig.10 Schematic diagram of the original image (left) and the image after sharpening and enhancement (right).
[0076] Fig.11 Schematic diagram of an image without corner detection and correction (top) and an image with corner detection and correction (bottom).
[0077] Fig.12 This is a schematic diagram of the carrier image (top) and the image after the ownership confirmation mark is embedded (bottom).
[0078] Fig.13 This is a comparison diagram of the visual quality between ASSW and SSR.
[0079] Fig.14 Table 1 shows the image quality comparison (SSIM) before and after corner detection of handheld images taken at different positions.
[0080] Fig.15 Table 2 shows the average error bits of the confirmation mark at different shooting distances.
[0081] Fig.16 Table 3 shows the average error bits of the confirmation mark at different angles in the horizontal direction.
[0082] Fig.17 Table 4 shows the average error bits of the confirmation mark at different angles in the vertical direction.
[0083] Fig.18 Table 5 shows the screen shot correction of random angles of handheld lenses. DETAILED DESCRIPTION
[0084] The present invention is further described in detail below in conjunction with the accompanying drawings.
[0085] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.
[0086] Embodiment 1:
[0087] A method for embedding an image authentication mark that resists screen shooting comprises the following steps:
[0088] Strength calculation: extract key points, use the absolute value of the Gaussian difference of the key points as the SIFT strength of the point, calculate the strength of each key point, sort them in descending order, and select the first n feature points as candidate feature points.
[0089] The key point is defined as p, and the intensity In(p) of the key point p is specifically defined as:
[0090] In(p)=|D(x,y,σ)|
[0091] Among them, D(x,y,σ) represents the Gaussian difference value of the key point, (x,y) is the coordinate in the image, and σ represents the variance of the Gaussian filter kernel.
[0092] Texture value calculation: Calculate the texture value of the 64×64 area around the candidate feature points, sort them in descending order, and use the preset number of feature points with the highest ranking as the final embedding center. The larger the texture value, the less obvious the visual quality of the embedded confirmation mark will be when embedding the title confirmation mark in the area. Therefore, it is necessary to perform a secondary screening on these n candidate points, calculate the texture value of the 64×64 area around them, sort them in descending order, and use the top five key points as the final embedding center.
[0093] The texture value is calculated by the AC coefficient of the regional DCT. The specific formula is as follows:
[0094]
[0095] After combining the regional texture value for filtering, the impact on image quality is alleviated to a certain extent compared to I-SIFT-based filtering. The visual effect is as follows: Figure 2 As shown in the picture on the right.
[0096] Sharpening: Sharpen the area around the feature point determined as the embedding center; the enhanced stability of the feature point can also ensure that the feature point can still be stably extracted after the image has undergone various disturbances, effectively improving the accuracy of locating the embedded area and increasing the success rate of extracting the title confirmation mark.
[0097] Image sharpening is a technique to enhance the edges and details in an image. The basic principle is to increase the intensity of the high-frequency components in the image. The sharpening operation is usually completed by convolution with a sharpening kernel (or filter). In this solution, a custom sharpening kernel is used, as shown in the following formula:
[0098]
[0099] Among them, α is a parameter that controls the sharpening strength. By convolving this kernel with the image, the local contrast of the image can be increased, thereby achieving a sharpening effect. In addition, in order to reduce the visual quality loss caused by image sharpening.
[0100] Weighted blending: After sharpening, choose to weight the original area and the sharpened area in a specific ratio. The specific formula is as follows:
[0101] I mixed =β·I s +(1-β)·I
[0102] Among them, β is a weight parameter between 0 and 1, which is used to control the sharpening area I s With the original region I in the final mixed image I mixed In this embodiment, β is set to 0.5.
[0103] The next step is to embed the ownership confirmation logo.
[0104] Figure 3 The complete process of embedding the ownership mark is shown. First, if the input image is in color, it will be converted to the YCbCr color space and only the Y channel will be selected as the carrier of the ownership mark. For grayscale images, it will be directly used as the carrier of the ownership mark. Once the carrier image is obtained, the next step involves SIFT feature point detection to obtain multiple potential key points.
[0105] Then, the key points are screened using the intensity-based SIFT algorithm and the texture complexity of the surrounding area to select the most suitable candidate areas for embedding the ownership confirmation mark. In these candidate areas, the ownership confirmation mark will first be processed by BCH encoding to ensure the reliability and integrity of the ownership confirmation mark information. Through this series of steps, an image containing the ownership confirmation mark is finally obtained.
[0106] After obtaining multiple embedding regions, an algorithm based on the size relationship of DCT coefficients is selected to embed the authentication mark. Past research has shown that this method has a certain effect in resisting screen shooting attacks, so the bits of the authentication mark can be stored based on this relationship.
[0107] The embedding process of the title confirmation mark includes:
[0108] Reconstruction of the confirmation identification sequence: The 64-bit 0 / 1 confirmation identification sequence is reconstructed into an 8×8 confirmation identification matrix. This process involves reorganizing the linear confirmation identification sequence into a matrix form for subsequent processing.
[0109] Split the embedding region and perform DCT transform: Select a 64×64 embedding region B in the image and split it into 64 8×8 sub-blocks. The purpose of this step is to subdivide the large embedding region into smaller blocks so that each block can be processed independently.
[0110] Select DCT coefficients: Apply discrete cosine transform to each 8×8 sub-block and select the intermediate frequency coefficients C1 and C2 at the positions (4,5) and (5,4) in each DCT block. They are chosen because they are less susceptible to image deformation while preserving image details.
[0111] Modify DCT coefficients: Apply discrete cosine transform to each 8×8 sub-block and select the intermediate frequency coefficients C1 and C2 at the two positions (4,5) and (5,4) in each DCT block. The specific modification rules are as follows:
[0112]
[0113] After the screen capture, the photo may be compressed by JPEG. During the compression process, the coefficients will be quantized to a certain extent. Therefore, it is necessary to ensure that the size relationship of the coefficients before and after quantization is consistent, that is:
[0114]
[0115] Among them, q1 and q2 are the quantization steps of the two positions (4,5) and (5,4) respectively. Represents a rounding down operation. In summary, the embedding formula of the entire property rights identification information is described as follows:
[0116]
[0117] In this process, r represents the embedding strength of the ownership confirmation mark, and r is not less than 1. In this study, the compression factor of JPEG is set to 50, and the values of q1 and q2 correspond to it.
[0118] Embedding: After adjusting the DCT coefficients using the above formula, an inverse DCT transform is performed on each DCT block of the image to complete the identification embedding process.
[0119] Repeat the embedding process: Repeat the above steps for the remaining 64×64 areas to be embedded until all embedded areas have completed the embedding of the ownership confirmation mark. This process ensures that the ownership confirmation mark information is embedded into the entire image to improve the robustness and reliability of the ownership confirmation mark.
[0120] Embodiment 2:
[0121] Figure 6The specific steps of the title confirmation mark extraction process are shown. First, SIFT feature point detection is performed on the carrier image. Then, a preliminary screening is performed based on the SIFT intensity of each feature point to obtain multiple candidate key points. Next, the texture value of the 64×64 area around these candidate key points is calculated, and the key points with the top ten texture values are selected as the final candidate points. In the title confirmation mark extraction process, a cross-validation-based method is used to determine the final title confirmation mark sequence, and finally BCH encoding is used for decoding to obtain the final title confirmation mark information. The following is a detailed introduction to the above key steps.
[0122] A method for extracting an image ownership identification that is resistant to screen shooting comprises the steps of:
[0123] Image acquisition: Acquire the title confirmation image to be extracted, wherein the title confirmation image to be extracted is obtained by reshooting the title confirmation image obtained by using the title confirmation image embedding method resistant to screen shooting as described in Example 1.
[0124] Image vertex confirmation: Receive the four blurred vertices of the target image in the screen shot image, use them as the center of the circle, and execute the Shi-Tomasi corner detection algorithm in a circular area with a radius of 10 pixels to find the accurate positions of the four vertices of the image.
[0125] Before the property rights identification mark is extracted, due to differences in shooting environment and camera parameters, different perspective distortions are easily produced in the screenshots. Therefore, after obtaining the photos, they need to be corrected first. Figure 4 This is the screen capture correction process, which is mainly divided into two steps: image corner detection and perspective distortion correction. Figure 5 shown.
[0126] Figure 5 In the method, four blurred vertices of the target image in the screen shot are manually selected, and the Shi-Tomasi corner detection algorithm is performed in a circular area with a radius of 10 pixels with them as the center, and finally the four accurate vertex positions of the image are found.
[0127] Image correction: After obtaining the coordinates of the four vertices of the image containing the property rights identification in the photo, use the following formula to correct its perspective distortion. Assume that the coordinates of the four vertices before correction are P1 (x1, y1), P2 (x2, y2), P3 (x3, y3) and P4 (x4, y4), and the coordinates of the four vertices after correction are P'1 (x'1, y'1), P'2 (x'2, y'2), P'3 (x'3, y'3) and P'4 (x'4, y'4). Substitute them into the following formulas:
[0128]
[0129] After obtaining the eight equations, we can solve them together to get eight variables: a0, a1, a2, b0, b1, b2, c1, c2. So far, we have obtained the mapping formula for the perspective distortion. The image in the photo is corrected by this formula, and finally a complete image with the property rights identification is obtained.
[0130] Key point confirmation: Select feature points within a preset range of intensity ranking in the image, calculate the regional texture values of these feature points, and take out feature points within a preset range of texture ranking as candidate key points. Specifically, the key point area screening is somewhat different from the embedding part of the title confirmation mark. The main difference is that the original feature area may change in the corrected carrier image. These changes may include three aspects: feature point position, SIFT intensity, and regional texture value. Although the title confirmation mark is repeatedly embedded in multiple areas, it cannot be completely guaranteed that their ranking will not shift as a whole, so it is necessary to increase the number of detected areas to ensure that the target feature points are covered.
[0131] In the regional screening during the embedding process, the top n feature points in SIFT intensity are selected. At the extraction end, the screening ranking is increased to 2n, which can ensure that the target is basically not missed in the initial screening. Secondly, the regional texture value may also change during the correction process. In the secondary screening, the regional texture values of the 2n feature points are calculated, and the top ten feature points in texture value are selected as the subsequent candidate key points.
[0132] Confirmation of the title confirmation area: In order to cope with the impact of the offset of the key point position, the error tolerance range of each key point is controlled within a small 3×3 area. After screening the last ten key points, the 3×3 area around each candidate key point is traversed to obtain the area that may contain the title confirmation mark, which is defined as the area to be extracted;
[0133] Extract the property rights identification.
[0134] The process of extracting the title confirmation mark includes:
[0135] The area to be extracted is divided into 8×8 sub-blocks, each of which is 8×8 pixels in size. The purpose of this step is to divide the entire area into smaller parts so that detailed frequency domain analysis can be performed in subsequent steps.
[0136] Perform DCT transform on the sub-blocks: Perform discrete cosine transform on each 8×8 sub-block. In the DCT coefficient matrix of these sub-blocks, select the intermediate frequency coefficients at positions (4,5) and (5,4), which are denoted as C1 and C2 respectively;
[0137] Extraction: After obtaining the intermediate frequency coefficients C1 and C2 of each sub-block, use the following formula to process these coefficients to extract the 8×8 ownership identification matrix. The specific content of the following formula may involve performing some operations on C1 and C2 to extract the ownership identification information hidden in these coefficients.
[0138]
[0139] Extract and flatten the title confirmation matrix of all regions: Repeat the above steps for all block regions to extract the 8×8 title confirmation matrix of each region, and then flatten these 8×8 title confirmation matrices in one dimension to obtain a linear title confirmation sequence. Finally, 10 sets of such title confirmation sequences can be obtained.
[0140] Screening the most similar pair of title confirmation identification: Compare the title confirmation identification sequences between different groups in pairs, and screen out the most similar pair of title confirmation identification sequences as candidate title confirmation identification pairs. The comparison method is to calculate the cosine similarity between the two title confirmation identification sequences, which can be the Pearson correlation coefficient or other similarity measurement standards. Finally, the most similar pair of title confirmation identification sequences is selected as a candidate pair to continue to perform subsequent steps.
[0141] Perform BCH decoding to obtain the final authentication mark: select any authentication mark sequence from the candidate authentication mark pair, perform BCH decoding operation, and thus restore the original authentication mark information. Through this process, the authentication mark information hidden in the image can be finally extracted.
[0142] Compared with the relevant existing technologies, the technical solutions proposed in the above embodiments 1 and 2 can maintain a high robust extraction accuracy rate when facing screen shooting attacks at respective angles. In terms of invisibility, the comparison before and after embedding the ownership confirmation mark can be referred to the figure.
[0143] By implementing the technical solution and comparing it with the SSR solution, the present invention has good resistance to screen capture attacks at various angles and shooting distances, as shown in Tables 2 to 5. At the same time, the invisibility of the ownership identification information of the present invention is also more advantageous, as shown in Tables 1, Fig.12 and Fig.13 shown.
[0144] Experimental method: In this experiment, the effective length of the property rights identification information is set to 39 bits. After BCH (64, 39) encoding, a coded sequence of length 64 is obtained, which can correct errors within 4 bits. The experiment selected 10 classic pictures from the USC-SIPI dataset for testing. The desktop monitor model used is "LG-27GP83B-B" and the mobile phone model used for shooting is "Redmi K60". We tested the regional sharpening enhancement and compared the changes in SIFT intensity values and visual quality before and after sharpening enhancement. The specific effects are as follows: Figure 8 , Fig. 9 and Fig.10 shown. Fig.11 The visual quality of the corrected image before and after corner detection is compared. The operating environment of this experiment is Windows 11 operating system and MATLAB R2018b. The experimental shooting environment and equipment are placed as follows Figure 7 shown.
[0145] In the first and second embodiments, the title confirmation mark is also called a watermark.
[0146] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "include..." or "comprise..." do not exclude the existence of other elements in the process, method, article or terminal device including the elements. In addition, in this article, "greater than", "less than", "exceed" and the like are understood to exclude the number itself; "above", "below", "within" and the like are understood to include the number itself.
[0147] The above description of the embodiments is to facilitate the understanding and use of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to the embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments. Improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the protection scope of the present invention.
Claims
1. A method for embedding a confirmation mark to resist screen shooting, characterized in that: Includes steps: Strength calculation: extract key points, use the absolute value of the Gaussian difference of the key points as the SIFT strength of the point, calculate the strength of each key point, sort them in descending order, and select the first n feature points as candidate feature points; Texture value calculation: Calculate the texture value of the 64×64 area around the candidate feature points, sort them in descending order, and use the preset number of feature points with the highest ranking as the final embedding center; Sharpening: sharpen the surrounding area of the feature point determined as the embedding center; Weighted blending: After sharpening, choose to weight the original area with the sharpened area in a specific ratio; The ownership confirmation logo is embedded.
2. According to the method for embedding a confirmation mark against screen shooting according to claim 1, it is characterized in that: The embedding process of the title confirmation mark includes: Reconstruction of the title confirmation sequence: Reconstruct the 64-bit 0 / 1 title confirmation sequence into an 8×8 title confirmation matrix; Split the embedding region and perform DCT transform: Select a 64×64 embedding region in the image and split it into 64 8×8 sub-blocks; Select DCT coefficients: Apply discrete cosine transform to each 8×8 sub-block and select the intermediate frequency coefficients C1 and C2 at the two positions (4,5) and (5,4) in each DCT block; Modify DCT coefficients; Embedding: Perform inverse DCT transform on each DCT block of the image to complete the identification embedding process; Repeat the embedding process: Repeat the above steps for the remaining 64×64 areas to be embedded until all the embedding areas have completed the embedding of the ownership confirmation logo.
3. According to the method for embedding an image authentication mark to resist screen shooting as described in claim 1, it is characterized in that: The key point is defined as p, and the intensity In(p) of the key point p is specifically defined as: In(p)=|D(x,y,σ)| Among them, D(x,y,σ) represents the Gaussian difference value of the key point, (x,y) is the coordinate in the image, and σ represents the variance of the Gaussian filter kernel.
4. According to the method for embedding an image authentication mark to resist screen shooting as described in claim 1, it is characterized in that: In the process of sharpening, the image is convolved with a sharpening kernel, which is as follows: Among them, α is a parameter that controls the sharpening strength.
5. An image authentication mark embedding system that resists screen shooting, characterized in that: include: Strength calculation unit: used to extract key points, take the absolute value of Gaussian difference of key points as the SIFT strength of the point, calculate the strength of each key point, sort them in descending order, and select the first n feature points as candidate feature points; Texture value calculation unit: used to calculate the texture value of the 64×64 area around the candidate feature points, sort them in descending order, and use the preset number of feature points with the highest ranking as the final embedding center; Sharpening unit: used to sharpen the surrounding area of the feature point determined as the embedding center; Weighted blending unit: used to select the original area and the sharpened area for weighted blending in a specific ratio after sharpening; The title confirmation logo embedding unit is used to embed the title confirmation logo.
6. A method for extracting image rights identification that is resistant to screen shooting, characterized in that: Includes steps: Image acquisition: Acquire the title confirmation image to be extracted, wherein the title confirmation image to be extracted is obtained by reshooting the title confirmation image obtained by using the image title confirmation image embedding method against screen shooting as described in any one of claims 1 to 4. Image vertex confirmation: Receive the four blurred vertices of the target image in the screen shot image, use them as the center of the circle, and perform the Shi-Tomasi corner detection algorithm in a circular area with a radius of 10 pixels to find the accurate positions of the four vertices of the image; Image correction: After obtaining the coordinates of the four vertices of the image containing the title confirmation mark in the photo, perform perspective distortion correction on it, and based on the result, correct the image in the photo to obtain a complete image containing the title confirmation mark; Key point confirmation: select feature points within a preset range of intensity ranking in the image, calculate the regional texture values of these feature points, and take out feature points within a preset range of texture ranking as candidate key points; Confirmation of the property rights identification area: Traverse the 3×3 area around each candidate key point to obtain the area that may contain the property rights identification, which is defined as the area to be extracted; Extract the property rights identification.
7. According to claim 6, a method for extracting image authentication marks that resist screen shooting is characterized in that: The process of extracting the title confirmation mark includes: Divide the area to be extracted into blocks, and divide the area of the title confirmation mark to be extracted into 8×8 sub-blocks, and the size of each sub-block is 8×8 pixels; Perform DCT transform on the sub-blocks: Perform discrete cosine transform on each 8×8 sub-block. In the DCT coefficient matrix of these sub-blocks, select the intermediate frequency coefficients at positions (4,5) and (5,4), which are denoted as C1 and C2 respectively; Extraction: After obtaining the intermediate frequency coefficients C1 and C2 of each sub-block, these coefficients are processed to extract an 8×8 identity matrix; Extract and flatten the property rights identification matrix of all regions: Repeat the above steps for all block regions to extract the 8×8 property rights identification matrix of each region. Then, flatten these 8×8 property rights identification matrices in one dimension to obtain a linear property rights identification sequence. Screening the most similar pairs of title confirmation identifiers: Compare the title confirmation identifier sequences between different groups in pairs, and screen out the most similar pair of title confirmation identifier sequences as candidate title confirmation identifier pairs; Perform BCH decoding to obtain the final authentication mark: select any authentication mark sequence from the candidate authentication mark pair, perform BCH decoding operation, thereby restoring the original authentication mark information and extracting the authentication mark hidden in the image.
8. According to claim 6, a method for extracting a confirmation mark that resists screen shooting is characterized in that: In screening the most similar pairs of title confirmation identifications, the comparison method is to calculate the cosine similarity between the two title confirmation identification sequences, and finally select the most similar pair of title confirmation identification sequences as candidate pairs to continue to execute subsequent steps.
9. According to claim 6, a method for extracting image authentication marks that resist screen shooting is characterized in that: The specific process of image correction includes: setting the coordinates of the four vertices before correction to P1(x1,y1), P2(x2,y2), P3(x3,y3) and P4(x4,y4), and the coordinates of the four vertices after correction to P'1(x'1,y'1), P'2(x'2,y'2), P'3(x'3,y'3) and P'4(x'4,y'4), and substituting them into the following formulas respectively: After obtaining the eight equations, the eight variables a0, a1, a2, b0, b1, b2, c1, and c2 are solved jointly, from which the mapping formula for perspective distortion is obtained. The image in the photo is corrected using this formula, and finally a complete image containing the property rights identification is obtained.
10. An image authentication mark extraction system that is resistant to screen shooting, characterized in that: include Image acquisition unit: used for acquiring the to-be-extracted ownership confirmation identification picture, wherein the to-be-extracted ownership confirmation identification picture is obtained by reshooting the ownership confirmation identification picture obtained by using the anti-screen-shooting image ownership confirmation identification embedding method as described in any one of claims 1-4. Image vertex confirmation unit: used to receive the four fuzzy vertices of the target image in the screen shot image, and use them as the center of the circle to perform the Shi-Tomasi corner detection algorithm in a circular area with a radius of 10 pixels to find the accurate positions of the four vertices of the image; Image correction unit: used to perform perspective distortion correction on the image containing the title confirmation mark in the photo after obtaining the coordinates of the four vertices of the image, and correct the image in the photo according to the result to obtain a complete image containing the title confirmation mark; Key point confirmation unit: used to select feature points within a preset range of intensity ranking in the image, calculate the regional texture values of these feature points, and take out feature points within the preset range of texture ranking as candidate key points; Title confirmation area confirmation unit: used to traverse the 3×3 area around each candidate key point to obtain the area that may contain the title confirmation mark, which is defined as the area to be extracted; The title confirmation identification extraction unit is used to extract the title confirmation identification.
Citation Information
Cited By
Screen-camera communication method for resisting light interference and imaging distortion
CN119743681A
