Anti-screening robust watermarking method based on harris-sift feature region

Abstract

The application belongs to the technical field of image watermarking, and discloses an anti-screening robust watermarking method based on a Harris-SIFT feature region, which comprises the following steps: using a Harris corner point detection algorithm based on entropy value weighting to extract feature points of an image, combining a SIFT algorithm to generate feature directions on the feature points, then constructing feature regions and screening out feature regions which do not overlap with each other for embedding of subsequent watermarking; generating watermarking information and embedding the feature regions; performing a camera shooting experiment on the watermarking image to perform a screen shooting attack on the watermarking image; performing geometric correction on the screen shooting attacked image to restore distortion of the image generated in the shooting process, and then extracting watermarking information from the corrected image after pretreatment. The method can resist screen shooting attacks and common image attacks, and is superior to current advanced anti-screening watermarking algorithms.

Description

Technical Field

[0001] This invention relates to the field of image watermarking technology, and in particular to a robust watermarking method against screen capture based on Harris-SIFT feature regions. Background Technology

[0002] With the widespread application of digital images and the popularization of the internet, the protection of digital image copyright has become increasingly important. To address this issue, digital watermarking technology has emerged. Digital watermarking is a technology that embeds digital information into multimedia data (such as images, audio, and video) to protect the copyright and integrity of digital content. Although digital watermarking technology has been widely applied in fields such as images, audio, and video, with technological advancements, the widespread adoption of smart mobile devices, and the improvement of chip technology, it is now possible to efficiently and effectively record information by taking photos of media files displayed on a mobile phone screen. Screen captures not only record valid information but also largely destroy the watermark signal, making the dissemination of these photos for information leakage more covert and difficult to trace. Traditional watermarking technologies can no longer meet the needs of watermark image information security; therefore, how to address the security problems caused by screen capture leaks has become an important research direction.

[0003] To address these issues, an increasing number of researchers have devoted themselves to the field of anti-photographic watermarking. Nakamura T and Katayama A et al. (Katayama, Atsushi, et al. "New high-speed frame detection method: Side trace algorithm (STA) for i-appli on cellular phones to detect watermarks." Proceedings of the 3rd international conference on Mobile andubiquitous multimedia. 2004) first discussed robust methods for print-photographic image watermarking. They proposed a spatial domain watermark embedding and detection scheme that extracts the watermark from the focus of an image captured by a camera-equipped mobile phone, using frame synchronization but only resisting minor geometric distortions. Pramila et al. (Pramila, A, Keskinarkaus A, Seppänen, T. Multiple domain watermarking for print-scan and JPEG resilientdata hiding. In Proceedings of International Workshop on Digital Watermarking; Springer: Berlin / Heidelberg, Germany, 2008; pp. 279–293) proposed a multi-domain-based algorithm for the print-scan process. They achieved robustness against geometric distortion by embedding a circular template in the discrete Fourier domain to correct for image rotation and scaling.

[0004] Based on the research of print-camera watermarking algorithms, Fang et al. (Fang H, Zhang W, Zhou H, et al. Screen-shooting resilient watermarking[J]. IEEE Transactions on Information Forensics and Security, 2018, 14(6): 1403-1418) proposed an intensity-based scale-invariant feature transform algorithm to address the special distortion problem in the screen shooting process. They combined a small template algorithm to repeatedly embed the watermark in different regions, which has high robustness and visual quality. However, when faced with simple textured images, this method cannot accurately locate the watermark region. Chen et al. (Chen W, Ren N, Zhu C, et al. Screen-Camrobust image watermarking with feature-based synchronization[J]. Applied Sciences, 2020, 10(21): 7494) proposed a feature-synchronization-based robust watermarking scheme for screen shooting to improve resilience to the severely distorted screen shooting process. They used an improved embedding algorithm to repeatedly embed messages into each selected local square feature region, demonstrating good robustness against screen-shooting and other common desynchronization attacks. Deng et al. (Deng B, Li S, Qian Z. An SVD-based screen-shooting resilientwatermarking scheme[J]. Multimedia Tools and Applications, 2022: 1-15) proposed a novel resilient screen-shooting watermarking scheme, utilizing the SIFT algorithm to locate the embedding region and embedding the watermark through the stability of the singular value matrix in terms of sign invariance. More regions were located during extraction to improve fault tolerance. Using this scheme, accurate watermark extraction was achieved under conditions of embedding more bits and better performance. Summary of the Invention

[0005] To address the shortcomings of existing anti-screen capture watermarking algorithms in terms of robustness and transparency, this invention proposes an anti-screen capture robust watermarking method based on Harris-SIFT feature regions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A robust watermarking method against screen capture based on Harris-SIFT feature regions includes:

[0008] Step 1: The Harris corner detection algorithm based on entropy weighting is used to extract feature points from the image, and the SIFT algorithm is combined to generate feature directions on the feature points. Then, feature regions are constructed and non-overlapping feature regions are selected for subsequent watermark embedding.

[0009] Step 2: Generate a 60-bit pseudo-random sequence as watermark information. After converting the feature region into the DFT domain, embed the watermark information into the feature region of the image using a circular template. After completing the watermark embedding of multiple feature regions, a complete watermarked image is obtained.

[0010] Step 3: Perform a screen capture attack on the watermarked image by taking a camera picture of the watermarked image.

[0011] Step 4: Perform geometric correction on the screen capture attack image to restore the distortion caused during the shooting process. After preprocessing the corrected image, extract the watermark information from it.

[0012] Furthermore, in step 1, the extraction of feature points from the image using the entropy-weighted Harris corner detection algorithm includes:

[0013] First, the image is preprocessed, including grayscale conversion and Gaussian filtering to extract local pixel value information and remove noise. Next, the entropy value is used as a weighting factor and multiplied with the Harris corner response function to obtain a new response function. Finally, non-maximum suppression is applied to the new response function value, and the maximum point obtained at the local location is the final corner point.

[0014] Furthermore, in step 1, the feature region is constructed in the following manner:

[0015] For each feature point, a feature region is constructed based on the feature scale, with the feature point as the center and the radius of the feature region as the radius. Defined as:

[0016]

[0017] in The constant coefficients, As a feature scale, Indicates rounding down;

[0018] After constructing a feature region for each feature point, the distance between feature points is limited, and feature regions with small feature scales are filtered out.

[0019] Further, step 2 includes:

[0020] Using keys Generate a pseudo-random sequence with a watermark of 60 bits;

[0021] Then, feature regions in the original carrier image are extracted sequentially, and the brightness band of the region is converted into the DFT domain;

[0022] The average grayscale value of the image is used as a reference to determine the watermark embedding radius. The average gray value of the feature region is calculated using the following formula:

[0023]

[0024] in and Let I(i,j) represent the width and height of the image, respectively, and let I(i,j) represent the grayscale value of the pixel located in the i-th row and j-th column of the image.

[0025] Calculate the watermark embedding radius based on the average grayscale value. :

[0026]

[0027] in It is a coefficient. It is the average gray value of the image;

[0028] According to the obtained watermark embedding radius The watermark information is embedded into the DFT amplitude of the feature region using the following formula:

[0029]

[0030] in Indicates the watermark width. Indicates embedding strength. The average value represents the amplitude. The standard deviation represents the amplitude;

[0031] The feature regions to be embedded with the watermark are restored into the brightness bands of square regions and converted into spatial domains to replace the corresponding regions in the original image. After all the feature regions to be embedded have been replaced, a watermarked image is finally obtained.

[0032] Furthermore, in step 4, the attacked image is geometrically corrected in the following manner to recover the distortions that occurred during the shooting process:

[0033] The transformation matrix is ​​determined by locating the four corner points of the embedded image in the screen-captured image and matching their corresponding coordinates with the coordinates in the corrected image. Then, the screen-captured image is transformed by perspective to obtain the corrected image.

[0034] Furthermore, in step 4, preprocessing the corrected image and then extracting the watermark information from it includes:

[0035] For the corrected image, the feature region is located. The feature region location process in the extraction stage is the same as the embedding process. First, a Gaussian function is performed on the corrected image to reduce the impact of noise attacks. Then, the Harris corner points and corresponding feature directions based on the entropy value of the image are calculated to obtain the feature regions. Finally, the corresponding feature regions are extracted for watermark message extraction.

[0036] The watermark message is detected and extracted sequentially for each feature region. When watermark information is detected in a feature region, the corresponding watermark extraction process is performed.

[0037] Furthermore, the watermark extraction process includes:

[0038] Perform on feature points The message sequence is obtained by traversing pixels to extract feature point locations and their neighborhood maximum values.

[0039]

[0040] in The extracted message sequence is defined. yes The maximum value extracted within the range, The average value is defined. The standard deviation is defined. It is a parameter used to determine the message extraction threshold.

[0041] Furthermore, after obtaining the message sequence, it also includes:

[0042] In message sequence After extraction, The watermark is compared with the original embedded pseudo-random sequence and the number of error bits is calculated. If the number of error bits is lower than the set threshold, the watermark extraction is successful.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) This invention proposes a Harris corner detection algorithm based on entropy weighting. By adaptively adjusting the size of the weighting factor, more robust corner points are extracted from different images, and points with low contrast and edge points are filtered out. This not only reduces the computation time, but also improves the accuracy and robustness of watermark detection.

[0045] (2) This invention proposes a watermarking algorithm based on a combination of Harris-SIFT feature regions and DFT transform. This algorithm performs DFT transforms on multiple selected feature regions and embeds the same watermark in each region. Even after an attack, as long as the watermark survives in one region, the watermark can be successfully extracted. This effectively improves the robustness of the watermark.

[0046] (3) This invention proposes a watermark radius selection method based on average gray value. The method of selecting the embedding radius based on the average gray value can adaptively select an appropriate embedding radius according to the content of different feature regions of the image, thereby improving the invisibility of the watermark.

[0047] (4) To verify the effectiveness of the present invention, a series of experiments were designed and conducted. The experimental results show that the algorithm of the present invention can resist screen capture attacks and common image attacks, and is superior to the current advanced anti-screen capture watermarking algorithms. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a robust watermarking method against screen capture based on Harris-SIFT feature regions according to an embodiment of the present invention.

[0049] Figure 2 This is an example diagram illustrating the formation of the watermark embedding area according to an embodiment of the present invention;

[0050] Figure 3 This is an example diagram of the correction process in an embodiment of the present invention;

[0051] Figure 4 These are example images of the original carrier image and the image after watermark embedding, as shown in this embodiment of the invention; the first row is the original image, and the second row is the image after watermark embedding.

[0052] Figure 5 Examples of Lena images taken at different distances according to embodiments of the present invention;

[0053] Figure 6 Examples of Lena images captured and extracted from different angles and distances according to embodiments of the present invention;

[0054] Figure 7 Examples of handheld images and corrected images are provided in this embodiment of the invention. Detailed Implementation

[0055] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:

[0056] A robust watermarking method against screen capture based on Harris-SIFT feature regions includes:

[0057] Firstly, in the feature point extraction stage, this invention proposes a Harris corner detection algorithm based on entropy weighting. By adaptively adjusting the weighting factor, more robust corner points are extracted from different images, while low-contrast points and edge points are filtered out. This not only reduces computation time but also improves the accuracy and robustness of watermark detection. Secondly, in the watermark embedding stage, this invention proposes a watermark radius selection method based on average grayscale values. This method, which selects the embedding radius based on the average grayscale value, adaptively selects an appropriate embedding radius according to the content of different feature regions in the image, thereby improving the invisibility of the watermark. Finally, in the watermark extraction stage, geometric distortion correction and Gaussian function preprocessing are performed on the captured image to extract the watermark information. The proposed anti-screen capture robust watermarking algorithm is as follows: Figure 1 As shown.

[0058] The specific technical solution of the present invention includes the following steps:

[0059] Step 1. Extract feature points from the image and construct feature regions: Use the improved Harris algorithm, namely the entropy-weighted Harris corner detection algorithm proposed in this invention, to extract feature points from the image, and combine it with the SIFT algorithm to generate feature directions on the feature points. Then construct feature regions and select non-overlapping feature regions for subsequent watermark embedding.

[0060] Step 2. Generate watermark information and embed it into the feature region: Generate a 60-bit pseudo-random sequence as watermark information. After converting the feature region into the DFT domain, embed the watermark information into the feature region of the image using a circular template. After completing the watermark embedding of multiple feature regions, a complete watermarked image can be obtained.

[0061] Step 3. Prepare experimental equipment to conduct camera shooting experiments on watermarked experimental images: Use a mobile phone camera to perform screen capture attacks on the images in a multi-angle, multi-distance and random shooting manner. Subsequently, extract the watermark information from the screen capture attacked experimental images and compare it with the embedded information to verify the anti-screen capture effect of the present invention.

[0062] Step 4. Extract watermarks from watermarked images subjected to screen capture attacks: Perform geometric correction on the attacked image to restore the distortion caused during the shooting process. After preprocessing the corrected image, extract the watermark information from it. Compare the extracted watermark information with the original watermark information to obtain the number of watermark error bits, which is used to analyze the robustness of the present invention.

[0063] This invention comprises two main parts: watermark embedding and watermark extraction. Watermark embedding is used to embed watermark information into experimental images in an advanced manner. Watermark extraction is responsible for extracting watermark information from the attacked screen capture image and verifying the accuracy of the extracted information.

[0064] The specific implementation schemes for realizing the present invention include the following embodiments:

[0065] 1. Watermark Embedding

[0066] First, the image is preprocessed, including grayscale conversion and Gaussian filtering to extract local pixel value information and remove noise. Next, the entropy value is used as a weighting factor and multiplied by the traditional Harris corner response function to obtain a new response function. This new response function can detect corners more accurately because it considers the degree and diversity of local pixel value changes. The formula for calculating the Harris corner response function is as follows:

[0067]

[0068] in for The determinant, For matrix traces, The response function of the corner point. The entropy is constant. This invention filters feature points with high entropy values ​​by calculating the entropy values ​​of pixels surrounding the keypoints. These feature points often indicate that the image region has more texture and structural information, exhibiting stronger robustness and stability, and are more likely to survive attacks. The feature point entropy value is multiplied by the Harris response function to obtain a new response function value. Finally, non-maximum suppression is applied to the new response function value. The local maxima obtained are the final corner points. Figure 2 (b) shows the Harris corners based on entropy weighting.

[0069] We calculate the feature orientation for each keypoint. We use the SIFT orientation descriptor to assign an orientation to each keypoint. The gradient magnitude and orientation are calculated as follows:

[0070]

[0071] in It refers to calculating the gray value at a specific point on an image.

[0072] By calculating the gradient magnitude and direction of keypoints, we can obtain the principal direction of each keypoint. For example... Figure 2 The arrows in (c) represent keypoints and their corresponding feature directions. For each keypoint, a feature region is constructed based on the feature scale. The radius of the feature region is calculated with the keypoint as the center. Defined as:

[0073]

[0074] in The constant coefficients, As a feature scale, This represents the floor function, which rounds down. Considering the severe distortion caused by screen capture attacks, the constructed feature regions should have sufficient range to ensure the survival of the watermark information. Therefore, after constructing a feature region for each feature point, the distance between feature points is limited, and feature regions with small scales are filtered out. Finally, feature regions of suitable size and non-overlapping features are obtained as the watermark embedding regions. Figure 2 (d) shows the feature region selected as the watermark embedding area.

[0075] Next, watermark embedding is performed. This is done using a key. Generate a pseudo-random sequence with a watermark of 60 bits. Then, feature regions in the original carrier image are extracted sequentially, and the brightness band of that region is converted to the DFT domain. This is to determine a suitable watermark embedding radius. This algorithm uses the average grayscale value of the image as a reference. The average grayscale value reflects the brightness of an image and can be applied to various types of images. The average grayscale value of the feature region can be calculated using the following formula:

[0076]

[0077] in and Let I(i,j) represent the width and height of the image, respectively, and let I(i,j) represent the grayscale value of the pixel located in the i-th row and j-th column of the image.

[0078] The watermark embedding radius is calculated based on the average grayscale value. The definition is as follows:

[0079]

[0080] in It is a coefficient. It is the average gray value of the image.

[0081] According to the obtained watermark embedding radius The watermark is embedded into the feature region. The watermark information is embedded into the DFT amplitude of the feature region using the following formula:

[0082]

[0083] in Indicates the watermark width. Indicates embedding strength. The average value represents the amplitude. The standard deviation of the amplitude is represented. The feature regions to be embedded with the watermark are restored into square brightness bands, and then converted to the spatial domain to replace the corresponding regions in the original image. After all the feature regions to be embedded have been replaced, a watermarked image is finally obtained.

[0084] 2. Watermark Extraction

[0085] Before extracting watermarks from attacked images, geometric distortion correction is necessary. Specifically, this involves locating the four corner points of the embedded image within the captured image, matching their coordinates with those in the corrected image to determine the transformation matrix, and then performing a perspective transformation on the original screen capture image to obtain the corrected image. Therefore, after extracting the captured image, we can locate the four vertices of the embedded image within the screen capture image. , , , and ,like Figure 3 As shown. Then we set the transformation coordinates corresponding to these 4 vertices as follows: , , and Substituting these 8 coordinates into the equation, we obtain 8 systems of equations, which, when solved, yield the following result: , , , , , , , The value of .

[0086]

[0087] After determining these parameters, a mapping from the distorted image to the corrected image can be established. Then, cropping and rescaling the corrected image generates the image to be extracted.

[0088] Feature regions are located in the corrected image. The feature region localization process in the extraction stage is the same as the embedding process. First, a Gaussian function is applied to the corrected image to reduce the impact of noise attacks. Then, Harris corner points based on entropy values ​​and corresponding feature directions are calculated to obtain feature regions. Finally, the corresponding feature regions are extracted for message extraction.

[0089] The watermark message is detected and extracted sequentially for each feature region. The watermark extraction process requires detecting each feature region; when watermark information is detected in a feature region, the corresponding watermark extraction process can begin. Because screen capture can slightly affect the watermark information, causing minor changes in the feature scale and its corresponding area, we will perform key point... The keypoint locations and their neighborhood maximum values ​​are extracted by traversing the pixels to obtain the message sequence.

[0090]

[0091] in The extracted message sequence is defined. yes The maximum value extracted within the range, The average value is defined. The standard deviation is defined. This is a parameter used to determine the message extraction threshold. In the message sequence... After extraction, the sequence is compared with the original embedded pseudo-random sequence and the number of error bits is calculated. If the number of error bits is below a certain threshold, the watermark extraction is successful.

[0092] 3. Experiment and Analysis

[0093] We evaluated the robustness of the watermark embedding algorithm through a series of experiments. The test images were sourced from Volume 3, Miscellaneous, of the USC-SIPI image database. We selected 14 color images plus a Lena image, for a total of 15 color images, and uniformly adjusted the image size during testing. For size testing, we chose the Bit Error Rate (BER), a commonly used metric for measuring the robustness of watermarking methods, as the evaluation standard. BER is defined as:

[0094]

[0095] in This refers to the number of error bits, which indicates the number of bits in the extracted watermark that do not match the original watermark information. A lower bit error rate (BER) means the extracted watermark is closer to the original watermark information, and also indicates that the watermarking algorithm has better robustness. In this experiment, the watermark detection threshold was set to 7, meaning that watermark detection was considered successful when the BER was less than 0.116.

[0096] In section 3.1, we conduct an invisibility test on the image with the embedded watermark. In section 3.2, we conduct a watermark robustness test on screen-captured images at different distances and angles. In section 3.3, we test the extraction performance of the watermarking algorithm when faced with random shooting.

[0097] 3.1 Invisibility Test

[0098] Figure 4 The image shows a visual quality comparison of four images—Lena, Mandril, Peppers, and House—after watermarking. The first row shows the original image, and the second row shows the watermarked image.

[0099] To verify the efficiency and reliability of this invention, an invisibility test was first conducted. Peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) are two commonly used metrics for evaluating the objective quality of images. PSNR measures the total pixel error between two images and is calculated as the logarithm of the ratio of pixel peak value to error.

[0100]

[0101] Where n is the pixel bit depth. For pixel peak value, This represents the mean square error.

[0102] SSIM is primarily used to measure the similarity between two images in terms of brightness, contrast, and structure. Its calculation formula is as follows:

[0103]

[0104] in and These are the means of images X and Y, respectively. and These are the standard deviations of images X and Y, respectively. The covariance of images X and Y, and These are constants related to pixel peak values. Higher PSNR and SSIM values ​​indicate less image distortion, with PSNR values ​​greater than 40dB indicating image quality very close to the original image. The objective quality of the 15 test images after watermark embedding is shown in the table below:

[0105] Table 1: Objective Quality of Watermarked Images

[0106]

[0107] According to the test results listed in Table 1, the watermarked images exhibit high quality in terms of PSNR and SSIM. Most images have a PSNR exceeding 41dB and an SSIM value above 0.96, indicating that the watermarked images maintain a high level of objective image quality. The overall average PSNR is 42.65, and the average SSIM is 0.9876, demonstrating that the watermarked images are very close to the original images.

[0108] 3.2 Robustness against screen capture attacks

[0109] In this section, we will verify the robustness of the watermarking algorithm against screen capture attacks. This invention is used to display watermarked images on screens with a resolution of [resolution missing]. A 24-inch monitor was used; photos were taken using a Mi 10s mobile phone. Images were displayed full-screen on the monitor, and focus and image quality were carefully controlled during shooting. Experimental data used... Figure 4 The four watermarked images shown illustrate the optimal shooting distance range for each image size, with moiré patterns appearing at certain distances. With the camera perpendicular to the screen, photos should be taken in 10cm increments at distances ranging from 30cm to 70cm. An example of this shooting method is shown below. Figure 5 As shown.

[0110] In a horizontal orientation, we set the shooting angle from vertical to 45° to the left / right of the horizontal, in 15° increments. Due to the influence of the shooting angle, the image could not be fully recorded when shooting from 30cm, so the shooting distance was set in 10cm increments, ranging from 40cm to 80cm. The captured and geometrically corrected Lena image is shown below. Figure 6 As shown.

[0111] Table 2: Average Bit Error Rate of Watermark Extraction at Different Angles and Distances

[0112]

[0113] Table 2 shows the average bit error rate (BER) of four experimental images under different shooting distances and angles. The data in the table shows that the shooting angle has a greater impact on the watermark than the shooting distance. When the horizontal shooting angle is within 30° to the left and right, and the distance is within 70cm, the average BER of the watermark is almost zero. With the increase of shooting distance and shooting angle, the BER of the watermark increases significantly, but overall it remains within the range where the watermark can be completely extracted. In summary, in practical applications, this invention can resist screen capture attacks in most cases. Therefore, this invention has strong robustness against screen capture attacks.

[0114] 3.3 Handheld Shooting Experiment

[0115] To more realistically simulate the randomness of screen capture, this invention also conducted experiments on watermark extraction from images under four random handheld shooting scenarios. Figure 7 Experimental results show that the watermarking algorithm proposed in this invention can successfully extract watermark information from images under different angles and shooting distances. This indicates that the algorithm of this invention has good applicability and robustness and can cope with various screen capture scenarios.

[0116] In summary, the watermarking algorithm proposed in this invention exhibits strong robustness and a low bit error rate under various shooting conditions, particularly demonstrating outstanding robustness at close to medium distances and different angles. This indicates that the watermarking algorithm proposed in this invention can be applied to a variety of different screen capture scenarios and has good practicality.

[0117] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A robust watermarking method against screen capture based on Harris-SIFT feature regions, characterized in that, include: Step 1: The Harris corner detection algorithm based on entropy weighting is used to extract feature points from the image, and the SIFT algorithm is combined to generate feature directions on the feature points. Then, feature regions are constructed and non-overlapping feature regions are selected for subsequent watermark embedding. Step 2: Generate a 60-bit pseudo-random sequence as watermark information. After converting the feature region into the DFT domain, embed the watermark information into the feature region of the image using a circular template. After embedding watermarks into multiple feature regions, a complete watermarked image is obtained. Step 3: Perform a screen capture attack on the watermarked image by taking a camera picture of the watermarked image. Step 4: Perform geometric correction on the image subjected to screen capture attack to restore the distortion caused during the shooting process. After preprocessing the corrected image, extract the watermark information from it. In step 1, the feature region is constructed in the following manner: For each feature point, a feature region is constructed based on the feature scale, with the feature point as the center and the radius of the feature region as the radius. Defined as: in The constant coefficients, As a feature scale, Indicates rounding down; After constructing a feature region for each feature point, the distance between feature points is limited, and feature regions with small feature scales are filtered out. Step 2 includes: Using keys Generate a pseudo-random sequence with a watermark of 60 bits; Then, feature regions in the original carrier image are extracted sequentially, and the brightness band of the region is converted into the DFT domain; The average grayscale value of the image is used as a reference to determine the watermark embedding radius. The average gray value of the feature region is calculated using the following formula: in and Let I(i,j) represent the width and height of the image, respectively, and let I(i,j) represent the grayscale value of the pixel located in the i-th row and j-th column of the image. Calculate the watermark embedding radius based on the average grayscale value. : in It is a coefficient. It is the average gray value of the image; According to the obtained watermark embedding radius The watermark information is embedded into the DFT amplitude of the feature region using the following formula: in Indicates the watermark width. Indicates embedding strength. The average value represents the amplitude. The standard deviation represents the amplitude; The feature regions to be embedded with the watermark are restored into the brightness bands of square regions and converted into spatial domains to replace the corresponding regions in the original image. After all the feature regions to be embedded have been replaced, a watermarked image is finally obtained.

2. The robust watermarking method against screen capture based on Harris-SIFT feature regions according to claim 1, characterized in that, In step 1, the feature point extraction of the image using the entropy-weighted Harris corner detection algorithm includes: First, the image is preprocessed, including grayscale conversion and Gaussian filtering to extract local pixel value information and remove noise. Next, the entropy value is used as a weighting factor and multiplied with the Harris corner response function to obtain a new response function. Finally, non-maximum suppression is applied to the new response function value, and the maximum point obtained at the local location is the final corner point.

3. The robust watermarking method against screen capture based on Harris-SIFT feature regions according to claim 1, characterized in that, In step 4, the attacked image is geometrically corrected in the following manner to restore the distortion that occurred during the shooting process: The transformation matrix is ​​determined by locating the four corner points of the embedded image in the screen-captured image and matching their corresponding coordinates with the coordinates in the corrected image. Then, the screen-captured image is transformed by perspective to obtain the corrected image.

4. The robust watermarking method against screen capture based on Harris-SIFT feature regions according to claim 1, characterized in that, Step 4, which involves preprocessing the corrected image and then extracting the watermark information from it, includes: For the corrected image, the feature region is located. The feature region location process in the extraction stage is the same as the embedding process. First, a Gaussian function is performed on the corrected image to reduce the impact of noise attacks. Then, the Harris corner points and corresponding feature directions based on the entropy value of the image are calculated to obtain the feature regions. Finally, the corresponding feature regions are extracted for watermark message extraction. The watermark message is detected and extracted sequentially for each feature region. When watermark information is detected in a feature region, the corresponding watermark extraction process is performed.

5. The anti-screen-capture robust watermarking method based on Harris-SIFT feature regions according to claim 4, characterized in that, The watermark extraction process includes: Perform on feature points The message sequence is obtained by traversing pixels to extract feature point locations and their neighborhood maximum values. in The extracted message sequence is defined. yes The maximum value extracted within the range, The average value is defined. The standard deviation is defined. It is a parameter used to determine the message extraction threshold.

6. The robust watermarking method against screen capture based on Harris-SIFT feature regions according to claim 5, characterized in that, After obtaining the message sequence, the following is also included: In message sequence After extraction, The watermark is compared with the original embedded pseudo-random sequence and the number of error bits is calculated. If the number of error bits is lower than the set threshold, the watermark extraction is successful.

Images