Digital image watermark registration positioning method and system based on linear Hough transform and application

Abstract

The invention discloses a digital watermark registration positioning method based on linear Hough transform, which is applied to positioning and identification of a digital watermark on a digital image, and comprises the following steps: encoding input data, adding random offset, generating basic unit watermark blocks, paving the basic unit watermark blocks, and reserving intervals for a linear detection positioning algorithm. In the extraction algorithm, firstly, preprocessing and a Hough straight line detection algorithm are carried out on a carrier image, detected straight lines are grouped, intersection point information between two groups is calculated, vertex positions of basic unit watermark blocks are positioned by calculating the mutual position relation between intersection points, and then perspective transformation correction is carried out on each basic unit watermark block. And the embedded basic unit watermark block information is obtained. And then extracting coding information from the basic unit watermark block, and sequentially carrying out mask removal, error correction and decoding to recover original embedded information. The invention further discloses a system and application for implementing the method, and the system and the application have wide application value.

Description

Technical Field

[0001] The present invention belongs to the technical field of image watermark registration and positioning, and relates to a registration and positioning method, system and application of digital watermark, which has high concealment, high recognition accuracy, strong error correction ability and confidentiality performance. Background Art

[0002] With the development of Internet technology, a large number of digital works are obtained and copied, and the security of their information and the problem of copyright protection have also become prominent. Digital watermark technology can encrypt and protect copyright information by embedding digital identification information in digital images without affecting the use value of the original carrier image.

[0003] In the prior art, digital watermark algorithms are mainly divided into two categories: spatial domain algorithms and transform domain algorithms. Since the robustness of transform domain algorithms is better than that of spatial domain algorithms, and the capacity of watermark information that can be embedded is larger, they have become the focus of current research on digital watermark technology. However, the transform domain algorithms are still not ideal in terms of the anti-attack ability and anti-extraction ability of watermarks, thus greatly limiting the application of digital watermark technology in other fields.

[0004] In the 1960s, the Hough line detection algorithm was proposed for analyzing tracks in a cloud chamber in particle physics. Later, this algorithm was extended to the field of computer vision and widely used in edge detection, image segmentation, object recognition and other fields.

[0005] In addition, traditional watermark technologies based on the idea of tiling divide the watermark image into equal blocks on average, so they fail to effectively utilize the characteristics (including affine transformation relationship and perspective transformation relationship) of the information between the watermark block images, and only divide the size of the embedded watermark block according to a custom size, without using the positional relationship between blocks. Summary of the Invention

[0006] In order to solve the deficiencies of the prior art, the purpose of the present invention is to provide a digital watermark registration and positioning method, system and application, that is, to determine information such as the size, position and direction of watermark blocks, including the methods of adding and extracting watermarks, based on the Hough line transform, so as to overcome the problem of low robustness of ordinary watermark algorithms after being attacked by perspective transformation, not only improve the recognition accuracy of watermarks, but also improve the robustness and concealment of watermarks, and effectively solve the disadvantages and limitations existing in the prior art.

[0007] The most core innovation of the present invention lies in tiling and embedding the basic unit of digital watermark, i.e., the watermark block, and leaving empty spaces for embedding straight lines between the basic unit watermark blocks. Thus, when extracting, the Hough line detection algorithm can be used to locate the vertex positions of the basic unit watermark blocks, and then perspective transformation correction is performed on each basic unit watermark block to obtain the information of the embedded basic unit watermark block.

[0008] The present invention provides a digital image watermark registration and positioning method based on the straight line Hough transform. The registration and positioning method includes two parts: watermark addition and watermark extraction;

[0009] The watermark addition includes the following steps:

[0010] Step 1: Read the information to be encoded, obtain the 256 - base data information according to the information to be encoded, and process the 256 - base data information. After encoding, generate a binary watermark;

[0011] The 256 - base data information includes the size of the basic unit watermark block, the encoded data information, and the error - correction information;

[0012] Convert the 256 - base data into binary data, and generate a binary mask information of the same length according to the length of the converted data;

[0013] According to the size of the basic unit watermark block, a mask corresponding to the size of the basic unit watermark block needs to be generated, and then the generated mask is used to add a random offset to the encoded data information and the error - correction information;

[0014] Specifically, convert the 256 - base data information into binary data information, perform an exclusive - OR operation with the generated mask to obtain a new binary data information with a random offset added (i.e., the binary watermark), and then place it in the data matrix of the size of the basic unit watermark block to obtain the finally embedded basic unit watermark block.

[0015] Step 2: Pre - process the original image to be watermarked, calculate the texture information and edge features of the pre - processed image, generate an adaptive embedding strength matrix, extract the BGR three - channel information of the image, splice the generated basic unit watermark blocks, and embed them into the image, and reconstruct the channels of the image to obtain the image after embedding the information;

[0016] Among them, the preprocessing of the original image includes adjusting the image size, grayscale value adjustment, filtering processing, etc. Adjusting the image size means cropping the picture, only cropping a part of the center of the original image. Generally, about 640 pixels by 640 pixels of the center of the original image can be reserved for registration and positioning through cropping; the grayscale value adjustment is adjusted according to the size of the average value of all pixels of the image. If the average value of the original image pixels is too large, a certain value is subtracted from all pixels of the original image to make the image darker, otherwise a certain value is added to make the image brighter. The increase and decrease of the pixel values of the original image are related to the exposure of the actual picture, and the values of different pictures will vary, making the overall grayscale of the image more balanced and avoiding overexposure or underexposure in the image; the filtering processing includes bilateral filtering, low-pass filtering, high-pass filtering, Wiener filtering, etc. Bilateral filtering can remove salt-and-pepper noise in the image, low-pass filtering can smooth the image and remove high-frequency noise, and the high-pass filter can enhance the edges and textures of the image; the Wiener filtering is used to minimize the estimation error caused by the introduction of noise, estimate the desired signal based on the mean square error of the original image, and is used for noise reduction and signal recovery, thereby improving the quality of the signal or image; the filtering processing can make the extracted features more obvious;

[0017] The generation process of the adaptive embedding intensity matrix is as follows: In the method proposed by the present invention, a weighting function is designed for the spatial domain of the image. The basic idea of the weighting function is to embed the watermark information into the less recognizable regions of the image with a higher weight, that is, the regions that are not easily recognizable, relatively similar in the image, and with insignificant differences. To achieve this, the image is divided into three regions: flat regions, strong edge regions, and texture regions. The judgment criteria for the regions are as follows:

[0018]

[0019] Aver(i,j), Std(i,j), and Edge(i,j) are the local average value, standard deviation, and edge detection value on x(i,j) respectively; x(i,j) represents the pixel value of the image. For edge detection, the Prewitt operator is used to determine the position of the edge by calculating the gradient of the pixel points in the image, and Aver Edge (I) and Std Edge (I) are the average value and standard deviation of the edge detection value Edge(i,j).

[0020] The weighting function of the above-designed image spatial domain is as follows:

[0021]

[0022] α is the minimum embedding strength, which is set to 2. WF(*) is a weighting function for dark and bright regions. Since dark and bright regions are less sensitive than normal regions, the watermark information is embedded with a higher weight through this function.

[0023] The result generated by the preprocessed image according to the weighting function is the adaptive embedding strength matrix.

[0024] The BGR three-channel information of the image includes: Blue channel (B): This channel contains the information of the blue component of each pixel in the image. The higher the value, the stronger the blue component; Green channel (G): This channel contains the information of the green component of each pixel in the image. The higher the value, the stronger the green component; Red channel (R): This channel contains the information of the red component of each pixel in the image. The higher the value, the stronger the red component.

[0025] When the basic unit watermark blocks are tiled and spliced, the pixel interval between every two basic unit watermark blocks is 1 pixel value, which is used for embedding a straight line for Hough line detection and positioning.

[0026] The original yellow-blue channel information I of the image is obtained by weighting the BGR three channels. B , the watermark information is w, the original image information is i, and the embedding strength matrix is the above-mentioned adaptive embedding strength matrix ρ, then I B ' = (1 - ρ) * I B + ρ * w; where I B ' represents the yellow-blue channel information of the image after embedding the information, I B represents the original yellow-blue channel information of the image, and ρ represents the embedding strength, i.e., the weight; the yellow-blue channel information of the image after embedding the watermark is obtained.

[0027] The B, G, and R channels of the reconstructed image are reconstructed, which means fusing the weighted I B ' channel information with the original B, G, and R channel information to generate an RGB color image, integrating the BGR three-channel information after embedding, and obtaining the image after embedding the information.

[0028] The watermark extraction includes the following steps:

[0029] Step I: Use an imaging recording device to obtain the image with the watermark to be extracted, and preprocess the image.

[0030] Step II: Calculate the texture information and edge features of the preprocessed image in Step I to generate an adaptive embedding strength matrix, and extract the yellow-blue channel information of the obtained image.

[0031] Step III: Use the inverse of the adaptive embedding strength matrix in Step II to extract the spliced watermark information in the yellow-blue channel.

[0032] Step IV: Use the Hough line detection algorithm to extract the positions of the lines in the stitched watermark information, divide the lines into two groups that are as vertical as possible, calculate the coordinate positions of the intersection points of the two groups of lines, and save them according to the line grouping.

[0033] Step V: Topologically expand outward from each intersection point to locate the positions of the four vertices of the upper left, lower left, upper right, and lower right corners of each block, and group them as one set. Perform perspective transformation correction on each set of four vertices to intercept the basic unit watermark blocks.

[0034] Specifically, during the watermark extraction process, scan the area on the original image through a mobile phone camera or other reading devices. Preprocess the obtained image with digital watermarking, including adjusting the image size, gray value adjustment, and filtering, to facilitate the extraction of digital watermarks in subsequent steps.

[0035] Calculate the texture information and edge features of the preprocessed image to generate an adaptive embedding intensity matrix.

[0036] Extract the BGR three-channel information of the image, weight the B, G, and R channels. The weighting function here is the same as that in the embedding part to obtain the yellow-blue channel information of the image. Use the inverse of the adaptive embedding intensity matrix to extract the stitched watermark information in the yellow-blue channel. The inverse extraction means calculating the difference between the original image and the preprocessed image, and combining with the adaptive embedding intensity matrix to obtain the embedded watermark information. Specifically, divide the difference by the above adaptive embedding intensity matrix to extract the watermark information.

[0037] Use the Hough line detection algorithm to perform threshold segmentation and binarization on the watermark image after inverse extraction. Generally, select the pixel average value of the watermark image after inverse extraction + 20 as the threshold. Since the embedded lines are generally white lines with the highest pixel value of 255, this threshold is sufficient to meet the requirements and there is still redundancy. Perform Hough line detection on the binarized image to obtain data of multiple lines, and extract the positions of the lines between adjacent basic unit watermark blocks in the stitched watermark information.

[0038] Use the clustering algorithm to divide the detected lines into two groups that are as vertical as possible. The clustering algorithms include k-means, K-Medoids, DBSCAN, Mean Shift, Agglomerative Hierarchical Clustering, which can divide multiple lines into multiple groups of lines. The angles of the lines within each group are similar, and the angles between groups are relatively large. The angles between different groups of lines are as close to 90° as possible.

[0039] The k-means clustering algorithm is relatively simple, intuitive, and efficient. It can converge quickly for large datasets and meet the requirements of this method. At the same time, calculate the coordinate positions of the intersection points of two groups of straight lines, group them by line, and divide the points on each line into a group and save them;

[0040] Using the relative position relationship between different intersection points, perform topology outward from each intersection point. The topology means that after obtaining the intersection points of straight lines, each intersection point belongs to two intersecting and approximately perpendicular straight lines. One can find the second intersection point along the first straight line where the current intersection point is located. At the same time, find the third intersection point along the second straight line where this intersection point is located. Calculate the difference between the relative distance x1 between the second intersection point and the first intersection point and the relative distance x2 between the third intersection point and the first intersection point. If the difference is less than 0.1*(x1 + x2), then it is considered that these three intersection points are the three vertices of the same basic unit watermark block. Otherwise, continue to find the next intersection point along the straight line until three points that meet the requirements are found. Then, the fourth point can be determined according to whether it is on the same straight line as the second intersection point, and at the same time on the same straight line as the third intersection point, and the difference between the distance x3 from the fourth point to the second intersection point and the distance x4 from the third intersection point is less than 0.1*(c3 + x4).

[0041] At this time, the positions of the four vertices, namely the upper left corner, lower left corner, upper right corner, and lower right corner, of each basic unit watermark block can be located and grouped;

[0042] Perform perspective transformation correction on each four-vertex group in the extracted watermark image, and after intercepting, obtain the basic unit watermark block to obtain the extraction information of the square watermark block.

[0043] The present invention also provides a system for implementing the above method. The system includes an encoding module, an embedding module, and an extraction module.

[0044] The encoding module further includes a data reading module, a random mask generation module, and a randomness offset module, and is used to perform the following operation steps:

[0045] (1) The data reading module reads the information to be encoded: it can be Chinese characters, English letters, URLs, symbols, numbers, etc.; among them, if the information to be encoded is a number, it can be directly encoded subsequently; if it is other (which can include Chinese characters, English letters, symbols, etc.) or a URL, it needs to be first converted to ASCII code values (ASCII code is a 256 - base number), and then encoded;

[0046] (2) The data reading module generates the following information: the size of the basic unit watermark block, the encoded data information, and the error correction information; the generated information is in the form of binary data converted from base-256 data; specifically, the information generated by the data reading module may include product information such as production date, product number, etc.; copyright information such as author name, publication time, etc.; image information, and the image information can be converted into digital information using tools.

[0047] (3) The random mask generation module generates a mask corresponding to the size of the basic unit watermark block, where the basic unit refers to the smallest image block required according to the amount of information to be embedded.

[0048] (4) The random offset module adds random offsets to the encoded data information and the error correction information using the generated mask, reducing the phenomenon that too many valid points in the generated basic unit watermark block are in the same row or the same column, and the arrangement pattern of the point set is obvious.

[0049] (5) The random offset module encodes to generate watermark information. The encoding information used by the random offset module in the process of generating the watermark specifically refers to the new binary data obtained after converting the base-256 data in (2) to binary and adding random offsets. The random offset refers to performing an exclusive OR operation on the binary data converted from the base-256 data in (2) and the mask generated by the random mask generation module in (3); the obtained watermark information is the new binary data, that is, the binary watermark.

[0050] (6) The random offset module places the binary watermark data generated in (5) into a data matrix of the size of the basic unit watermark block to obtain the basic unit watermark block.

[0051] The embedding module further includes a preprocessing module, an adaptive embedding intensity matrix module, and a weighting module, which can embed the generated basic unit watermark block containing binary watermark information into the original image; and is used to perform the following steps:

[0052] (1) The preprocessing module preprocesses the original image that needs to embed the watermark, including adjusting the image size, grayscale value adjustment, filtering processing, etc.; adjusting the image size means cropping the picture, only cropping a part of the center of the original image for registration and positioning; the grayscale value adjustment is adjusted according to the size of the average value of all pixels in the image. If the average value of the original image pixels is too large, a certain value is subtracted from all pixels of the original image to make the image darker, otherwise a certain value is added to make the image brighter; the filtering processing includes bilateral filtering, low-pass filtering, high-pass filtering, Wiener filtering, etc. Bilateral filtering can remove salt-and-pepper noise in the image, low-pass filtering can smooth the image and remove high-frequency noise, while the high-pass filter can enhance the edges and textures of the image; the Wiener filtering is used to minimize the estimation error caused by noise introduction, thereby improving the quality of the signal or image. Filtering processing can make the extracted features more obvious, facilitating the embedding of digital watermarks in subsequent steps;

[0053] (2) The adaptive embedding strength matrix module calculates the texture information and edge features of the preprocessed image to generate an adaptive embedding strength matrix; in the proposed method, a weighting function is designed for the spatial domain of the image. The basic idea of the weighting function is to embed the watermark information with a higher weight into the less recognizable areas of the image. To achieve this, the image is divided into three regions: flat region, strong edge region, and texture region. The judgment criteria for the regions are as follows:

[0054]

[0055] Aver(i,j), Std(i,j), and Edge(i,j) are the local average value, standard deviation, and edge detection value on x(i,j) respectively. For edge detection, the Prewitt operator is used to determine the position of the edge by calculating the gradient of the pixel points in the image, and Aver Edge (I) and Std Edge (I) are the average value and standard deviation of the edge detection value Edge(i,j).

[0056] The weighting function is as follows:

[0057]

[0058] α is the minimum embedding strength, set to 2. WF(*) is a weighting function for dark and bright areas. Since dark and bright areas are less sensitive than normal areas, the watermark information is embedded with a higher weight through this function.

[0059] (3) The weighting module extracts the BGR three-channel information, splices the basic unit watermark blocks, and weights the B, G, and R channels to obtain the yellow and blue channel information of the image after embedding the watermark;

[0060] I B ′ = α × B - β × R - γ × G

[0061] Among them, α, β, and γ are weighting coefficients, which can be adjusted as needed. The default value of α is 1, and the default values of β and γ are 0.5.

[0062] The BGR three-channel information includes: Blue channel (B): This channel contains the information of the blue component of each pixel in the image. The higher the value, the stronger the blue component; Green channel (G): This channel contains the information of the green component of each pixel in the image. The higher the value, the stronger the green component; Red channel (R): This channel contains the information of the red component of each pixel in the image. The higher the value, the stronger the red component.

[0063] The weighted module tiles and splices the generated basic unit watermark blocks. At the same time, the pixel interval between every two basic unit watermark blocks is 1 pixel value, which is used to embed a straight line for Hough line detection and positioning, and to embed straight line positioning information.

[0064] The B, G, and R three channels are weighted to obtain the original yellow-blue channel information I of the image B , the watermark information is w, the original image information is I, and the adaptive embedding strength matrix is the above-mentioned adaptive embedding strength matrix ρ. Then I B ' = (1 - ρ) * I B + ρ * w; where I B ' represents the yellow-blue channel information of the image after embedding the information, I B represents the yellow-blue channel information of the original image, and ρ represents the embedding strength, that is, the weight; obtain the yellow-blue channel information of the image after embedding the watermark.

[0065] The extraction module further includes a preprocessing module, an adaptive embedding strength matrix module, an inverse extraction module, a straight line detection module, and a registration and positioning module; and is used to perform the following steps:

[0066] (1) The preprocessing module scans the part on the original image through a mobile phone camera or other reading devices, and preprocesses the obtained image embedded with the digital watermark, including adjusting the size of the image, adjusting the gray value, and filtering, so as to facilitate the extraction of the digital watermark in the subsequent steps.

[0067] (2) The adaptive embedding strength matrix module calculates the texture information and edge features of the preprocessed image, and generates an adaptive embedding strength matrix.

[0068] (3) The BGR three-channel information of the image is extracted by the inverse extraction module, and the B, G, and R channels are weighted. The weighting function here is the same as that in the embedding part; the yellow-blue channel information of the image is obtained, and the spliced watermark information in the yellow-blue channel is inversely extracted by using the adaptive embedding intensity matrix. The inverse extraction refers to calculating the difference between the original image and the preprocessed image, and the embedded watermark information can be obtained by combining the adaptive embedding intensity matrix. Specifically, the watermark information is extracted by dividing the difference by the above adaptive embedding intensity matrix;

[0069] (4) The straight line detection module uses the Hough straight line detection algorithm to perform threshold segmentation and binarization on the watermark image after inverse extraction. Generally, the pixel average value of the watermark image after inverse extraction + 20 is selected as the threshold. Since the embedded straight line is generally a white line with the highest pixel value of 255, this threshold is sufficient to meet the requirements and there is still redundancy. The Hough straight line detection is performed on the binarized image to obtain the data of multiple straight lines, and the positions of the straight lines between adjacent basic unit watermark blocks in the spliced watermark information are extracted;

[0070] (5) The straight line detection module uses the clustering algorithm to divide the detected straight lines into two groups that are as perpendicular as possible. The clustering algorithm can divide multiple straight lines into multiple groups of straight lines. The angles of the straight lines within each group are similar, and the angles between groups are quite different;

[0071] The general algorithm requirements of k-means include: 1. The number of clusters needs to be determined in advance; 2. Select the initial center; 3. Distance measurement; 4. Sensitive to outliers; 5. Line boundaries.

[0072] In addition to the k-means clustering algorithm, there are also clustering algorithms such as K-Medoids, DBSCAN, Mean Shift, and Agglomerative Hierarchical Clustering. k-means is relatively simple and intuitive, and has high efficiency and can converge quickly for large datasets, which can meet the needs of this method. At the same time, calculate the coordinate point positions where the two groups of straight lines intersect, group them by line, and divide the points on each line into a group and save them;

[0073] (6) The registration and positioning module uses the relative position relationships between different intersection points to perform topology outward from each intersection point. Here, "topology" means that after obtaining the intersection points of lines, each intersection point belongs to two intersecting and approximately perpendicular lines. Starting from the current intersection point, the second intersection point can be found along the first line where it is located, and at the same time, the third intersection point can be found along the second line where it is located. Calculate the difference between the relative distance x1 between the second intersection point and the first intersection point and the relative distance x2 between the third intersection point and the first intersection point. If the difference is less than 0.1*(x1 + x2), it is considered that the three intersection points are the three vertices of the same basic unit watermark block; otherwise, continue to find the next intersection point along the line where it is located until three points that meet the requirements are found. Then, the fourth point can be determined based on whether it is on the same line as the second intersection point, on the same line as the third intersection point, and the difference between the distance x3 from this point to the second intersection point and the distance x4 from this point to the third intersection point is less than 0.1*(x3 + x4).

[0074] At this time, the positions of the four vertices, namely the upper left corner, lower left corner, upper right corner, and lower right corner, of each basic unit watermark block can be located and grouped as a set.

[0075] (7) The registration and positioning module performs perspective transformation and correction on each four-vertex group in the extracted watermark image, and after cropping, the basic unit watermark block is obtained, and the extraction information of the square watermark block is obtained.

[0076] The present invention also provides the applications of the above registration and positioning method or registration and positioning system in copyright protection, content authentication and verification, broadcast monitoring, fingerprint recognition, etc., as follows:

[0077] 1. Copyright protection: Add watermarks to images, audio, or videos to identify the owner or creator of the content. When the content is illegally copied or distributed, the source can be identified and traced by extracting the watermark.

[0078] 2. Content authentication and verification: Ensure the integrity and authenticity of digital media. For example, medical images or court evidence may require watermarks to ensure that they have not been tampered with.

[0079] 3. Broadcast monitoring: Television and radio companies can embed watermarks in their content to monitor the location and time of their broadcasts. This helps to track the distribution and broadcast of the content.

[0080] 4. Fingerprint recognition: Generate unique watermarks for each user or device and embed them into the distributed content. If the content is illegally shared, the source of the leak can be identified through the watermark.

[0081] Due to the adoption of the above technical solutions, the present invention brings the following beneficial effects:

[0082] The image registration and positioning algorithm of the present invention has high robustness to perspective transformation and radiation transformation, and the digital watermark blocks can be easily positioned without positioning points, eliminating the requirement for positioning points of ordinary digital watermark blocks. The same watermark blocks can be repeatedly printed and spliced into a watermark block array, and it is convenient to obtain watermark blocks; it simplifies the method of adding positioning points to the basic watermark blocks, making the basic unit watermark blocks small and concise. The unit digital watermark blocks can be as small as 64*64 at least, saving packaging space while improving the coding efficiency. The straight lines embedded in the present invention do not need to be clearly visible to the naked eye and do not need to occupy blank space alone, so they will not affect the layout of the image and are convenient and flexible to use.

[0083] The following are the existing technologies applied to similar scenarios, and the relevant advantages and disadvantages are as follows:

[0084] 1. SIFT: Advantage: It is invariant to scale, rotation, and partial perspective changes. It has good distinctiveness. Disadvantage: The calculation is relatively slow and may be restricted by patents.

[0085] 2. SURF: Advantage: Faster than SIFT and invariant to scale and rotation. Disadvantage: It may not be as stable as SIFT.

[0086] 3. ORB: Advantage: Very fast and useful for real-time applications. Disadvantage: It may not be as stable as SIFT or SURF.

[0087] 4. Mutual information: Advantage: Suitable for multimodal image registration and does not depend on the intensity of the image. Disadvantage: The calculation is complex and an optimization algorithm is required.

[0088] 5. Normalized cross-correlation: Advantage: The result is intuitive and suitable for template matching. Disadvantage: Sensitive to noise.

[0089] 6. Fourier transform: Advantage: Fast and suitable for global registration. Disadvantage: It may not be applicable to local or non-linear transformations.

[0090] 7. Histogram-based registration method: Advantage: Simple calculation and robust to brightness and contrast changes. Disadvantage: It may not be applicable to complex scenarios or multimodal images.

[0091] 8. Deep learning method: Advantage: It can learn complex transformations and may be more accurate than traditional methods for certain tasks. Disadvantage: A large amount of labeled data is required for training and high computing resources are required.

[0092] The image registration and positioning algorithm of the present invention is very simple, the data reading is fast, and it has high robustness and can detect simple geometric shapes, even in the case of noise or partial occlusion. However, it may not be accurate enough for complex shapes.

[0093] The image registration and positioning algorithm of the present invention has a high fault tolerance rate. The straight line detection algorithm can achieve pixel-level detection accuracy by adjusting parameters. Therefore, even if there is a small amount of damage or occlusion on the straight line position, normal matching and positioning can still be carried out. At the same time, multiple vertices will be intersected by the two groups of detected straight lines. Multiple groups of basic unit watermark blocks can be constructed through topological traversal and positional relationships. By selecting a group of intact basic unit areas in different regions without damage and occlusion for identification, the robustness of this method is greatly increased. At the same time, the amount of stored information is flexible, and the size and fault tolerance rate of the basic unit watermark blocks can be adjusted according to the storage requirements. When the data storage requirement increases, the size of the required basic unit watermark blocks also expands accordingly. Similarly, in order to improve the fault tolerance, more error correction codes must be introduced, which will also lead to an increase in the size of the basic unit blocks. Conversely, when the storage requirement or fault tolerance rate decreases, the size of the basic unit blocks will also decrease accordingly. Therefore, based on different storage requirements and fault tolerance criteria, the required size of the basic unit blocks will be different. At the same time, the intensity can be adaptively adjusted and the embedding intensity can be adjusted, with high concealment. Each basic unit watermark block represents one bit of valid information by a single pixel and is almost invisible to the naked eye when embedded in the image, which can effectively reduce the possibility of being damaged, copied, and forged. It has strong anti-perspective transformation ability. Since the four vertex positions of each basic unit watermark block can be obtained, the accuracy is relatively high when performing inverse perspective transformation restoration, effectively avoiding the problem that the traditional digital watermark algorithm has weak horizontal resistance to affine transformation attacks.

[0094] The present invention has a wide range of application scenarios and can be used for anti-counterfeiting, traceability, information hiding, etc. The image registration and positioning algorithm of the present invention only needs to be embedded in the yellow and blue channels. Because the human eye is not sensitive to the color transformation of the yellow and blue channels, it has high invisibility. As long as there is a sufficient difference between the brightness value embedded in the yellow and blue channels and the brightness value of the background color, the readability of the point cloud code can be guaranteed. The background color can directly use the background color of paper or other printing media, saving production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative efforts.

[0096] Figure 1 It is a flowchart of digital watermark encoding.

[0097] Figure 2 It is a flowchart of digital watermark embedding.

[0098] Figure 3 It is a flowchart of digital watermark extraction.

[0099] Figure 4 This is the flowchart of the embedding method of the digital image watermark registration and positioning algorithm based on the straight-line Hough transform in the present invention.

[0100] Figure 5 This is the flowchart of the extraction method of the digital image watermark registration and positioning algorithm based on the straight-line Hough transform in the present invention. Specific embodiments

[0101] In combination with the following specific embodiments and drawings, the present invention will be further described in detail. The processes, conditions, experimental methods, etc. for implementing the present invention are all common knowledge and well-known common sense in the art except for the specifically mentioned content below, and the present invention has no special limitations.

[0102] The present invention discloses a digital image watermark registration and positioning method based on the straight-line Hough transform. This method is mainly applied to the positioning and recognition of digital watermarks on digital images. By encoding the input data and adding random offsets, basic unit watermark blocks are generated. The basic unit watermark blocks are tiled, and intervals are reserved for the straight-line detection and positioning algorithm. In the extraction algorithm, first, the carrier image is preprocessed and the Hough straight-line detection algorithm is used. The detected straight lines are divided into two approximately perpendicular groups, and the intersection information between the two groups is calculated. By calculating the mutual positional relationship between the intersections, the vertex positions of the basic unit watermark blocks are located, so as to perform perspective transformation correction on each basic unit watermark block to obtain the embedded basic unit watermark block information. Then, the encoded information is extracted from the basic unit watermark blocks, and demasking, error correction, and decoding are performed in sequence to restore the original embedded information. Among them, the present invention mainly uses the Hough straight-line detection algorithm to achieve the positioning, perspective correction, and extraction of digital watermarks in images. This algorithm has high precision, high efficiency, and robustness, and can be applied to the positioning and extraction of various digital watermarks. This algorithm can effectively improve the recognition accuracy and robustness of digital watermarks, and has low requirements for reading devices, and has broad application value.

[0103] Specifically,

[0104] The present invention provides a digital image watermark registration and positioning method based on the straight-line Hough transform to overcome the problem that ordinary digital watermark algorithms are weak in resisting affine transformation attacks and perspective transformation attacks. The registration and positioning method includes the encoding and embedding of watermarks and the extraction of watermarks;

[0105] The encoding part includes the following operation steps:

[0106] 1. Read the encoded information

[0107] Read the input information through the device. The input data can be Chinese characters, English letters, URLs, symbols, numbers, etc. Among them, if it is a number, it is directly encoded; if it is a text (including Chinese characters, English letters, etc.), symbol or URL, it needs to be converted to its ASCII code value first and then encoded.

[0108] II. Generate corresponding information

[0109] Generate the required basic unit watermark block size, encoded data information and error correction information according to the read data. It should be noted that the data reading module can determine the required encoded data length according to the length of the read data and the user's fault tolerance requirements, and select an appropriate basic unit watermark block size. When the data storage requirement increases, the size of the required basic unit watermark block also expands accordingly. Similarly, in order to improve fault tolerance, more error correction codes must be introduced, which will also lead to an increase in the size of the basic unit block. Conversely, when the storage requirement or fault tolerance rate decreases, the size of the basic unit block will also decrease accordingly. Therefore, based on different storage requirements and fault tolerance standards, the required basic unit block size will vary.

[0110] III. Add a mask

[0111] Generate a mask corresponding to the basic unit watermark block size by the mask generation module; the mask is randomly generated and will be stored in the decoding system as the key for inverse masking during decoding;

[0112] IV. Add random offsets

[0113] The random offset module adds random offsets to the mask, encoded data information, and error correction information to reduce the phenomenon that too many valid points in the generated basic unit watermark block are in the same row or the same column and the arrangement rule is obvious;

[0114] V. Encoding

[0115] Generate the corresponding encoded information from the information added with random offsets and masks. The encoded information is binary data;

[0116] The described embedding device performs the following operating steps:

[0117] I. Image preprocessing

[0118] Preprocess the original image that needs to embed the watermark, including adjusting the size of the image to meet the requirements of the image and performing filtering processing on the gray value to facilitate embedding the digital watermark in the subsequent steps;

[0119] The bilateral filtering can remove the salt-and-pepper noise in the image, the low-pass filtering can smooth the image and remove the high-frequency noise, while the high-pass filter can enhance the edges and textures of the image; the Wiener filter is a linear filter with the minimum square as the optimal criterion, which is a filter that minimizes the square of the difference between the output and a given function (usually called the desired output) under certain constraints;

[0120] II. Generate the embedding intensity matrix

[0121] Calculate the texture information and edge features of the preprocessed image to generate an adaptive embedding intensity matrix;

[0122] III. Extract the yellow-blue channel information

[0123] Extract the BGR three-channel information of the image, and perform weighting on the B, G, and R channels to obtain the yellow-blue channel information of the image. The extraction formula is as follows:

[0124] b = (102 / 255)R + (255 / 255)G + (153 / 255)B

[0125] where b is the extracted yellow-blue channel information;

[0126] IV. Stitch the unit watermark blocks

[0127] Tile and stitch the generated basic unit watermark blocks, and leave a unit pixel interval between every two blocks and set the pixels to 255 to embed the line positioning information;

[0128] V. Embed the watermark information

[0129] A weighting function is designed for the spatial domain of the image. The basic idea of the weighting function is to embed the watermark information into the less recognizable regions of the image with a higher weight. To achieve this, the image is divided into three regions: flat region, strong edge region, and texture region. The judgment criteria for the regions are as follows:

[0130]

[0131] Aver(i,j), Std(i,j), and Edge(i,j) are the local average, standard deviation, and edge detection value on x(i,j) respectively. For edge detection, the Prewitt operator is used, and Aver Edge (I) and std Edge (I) are the average and standard deviation of the edge detection value Edge(i,j). The weighting function is as follows:

[0132]

[0133] α is the minimum embedding strength, which is set to 2. WF(*) is a weighting function for dark and bright regions. Since these regions (dark and bright regions) are less sensitive than normal regions, the watermark information is embedded with a higher weight through this function.

[0134] The embedding coefficient is designed using an adaptive embedding strength matrix, and the spliced watermark block is embedded into the yellow and blue channels of the image. The embedding formula is as follows:

[0135] I B ′ = (1 - p) * I B + ρ * w

[0136] where ρ is the embedding strength matrix, w is the digital watermark information, and I B is the original yellow and blue channel information of the image, and I B ' represents the yellow and blue channel information of the image after embedding the information;

[0137] VI. Generating the watermarked image

[0138] Reconstruct the B and G channels of the image, integrate the BGR three-channel information after embedding, and obtain the image after embedding the information.

[0139] The described extraction device performs the following steps:

[0140] I. Preprocessing of the scanned image

[0141] Scan the area on the image through a mobile phone camera or other reading devices, preprocess the obtained image with embedded digital watermark, adjust the size of the image, and perform Wiener filtering on the grayscale value to facilitate the extraction of the digital watermark in the subsequent steps;

[0142] II. Generating the embedding strength matrix

[0143] Calculate the texture information and edge features of the preprocessed image to generate an adaptive embedding strength matrix;

[0144] III. Extracting the yellow and blue channel information

[0145] Extract the BGR three-channel information of the image, and weight the B, G, and R channels to obtain the yellow and blue channel information of the image;

[0146] IV. Extracting the watermark information

[0147] Use the inverse of the adaptive embedding strength matrix to extract the spliced watermark information in the yellow and blue channels. The extraction formula is as follows:

[0148] w = (c - b) / α + b

[0149] Among them, α is the embedding strength matrix, w is the digital watermark information, b is the yellow-blue channel information of the image, and c is the image after Wiener filtering;

[0150] V. Straight Line Hough Detection

[0151] Use the straight line Hough detection algorithm to identify the binarized watermark information image, extract the polar coordinate positions (ρ, θ) of the straight lines in the image, and save them into the straight line information group;

[0152] VI. Calculate Intersection Points

[0153] Use the k-means clustering algorithm to divide the straight line information group into two groups of straight lines that are as perpendicular as possible, calculate the intersection points of the two groups of straight lines with each other, then group by line, and divide the points on each line into a group and save them. The intersection point calculation formula is:

[0154]

[0155] Among them, (x0, y0) are the intersection point coordinates, and (ρ1, θ1), (ρ2, θ2) are the polar coordinates of the two straight lines;

[0156] VII. Topological Traversal

[0157] Use the relative position relationship between points to perform topological traversal outward from each point, locate the positions of the upper left, lower left, upper right, and lower right vertices of each block, divide them into a group, and save the information group;

[0158] VIII. Extract Watermark

[0159] Perform perspective transformation correction on each four-vertex group in the extracted watermark image, and intercept to obtain the basic unit watermark block.

[0160] Embodiment 1

[0161] Taking copyright protection as an example, the method of this embodiment adds watermarks to images, audio, or videos to identify the owner or creator of the content. When the content is illegally copied or distributed, the source can be identified and traced by extracting the watermark. The entire process and implementation of this method specifically include:

[0162] 1. Watermark Encoding: Encode the copyright information or identifier into 256 - base data information, and process the 256 - base data information. After encoding, generate a binary watermark;

[0163] 2. Generate an Adaptive Embedding Strength Matrix: Pre - process the original image to be embedded with the watermark, calculate the texture information and edge features of the pre - processed image, and generate an adaptive embedding strength matrix;

[0164] 3. Watermark embedding: Extract the yellow and blue channel information of the image, splice the generated basic unit watermark blocks, and leave a 1-pixel space between the basic unit watermark blocks for embedding straight lines. According to the generated adaptive embedding strength matrix, embed it into the image, reconstruct the channels of the image, and obtain the watermarked image after embedding the information;

[0165] 4. Watermark detection: When it is necessary to detect or extract the watermark, generate the adaptive embedding strength matrix again to process the image that may contain the watermark. By comparing the original image and the image after Wiener filtering, detect the watermark information using the adaptive embedding strength matrix;

[0166] 5. Watermark extraction: Use the Hough line detection algorithm to extract the positions of the lines in the spliced watermark information, and divide the lines into two groups that are as vertical as possible. Calculate the coordinate positions of the intersection points of the two groups of lines, save them by line grouping. At the same time, calculate the positions where the lines of the two groups generate intersection points, topologically expand outward from each point, locate the four vertex positions of the upper left corner, lower left corner, upper right corner, and lower right corner of each block as a group, and perform perspective transformation correction on each four-vertex group, intercept the basic unit watermark block and decode it to obtain the original copyright information or identifier;

[0167] 6. Verification and tracing: Once the watermark is successfully extracted, its content can be verified to confirm the source of the image, audio or video.

[0168] If illegal copying or distribution is detected, the extracted watermark information can be used to trace and identify the source.

[0169] Comparative example

[0170] The watermark embedding and extraction method based on SIFT (Scale-Invariant Feature Transform) specifically includes the following steps:

[0171] 1. Feature extraction: Use the SIFT algorithm to process the original image and extract the key points and their descriptors.

[0172] 2. Watermark encoding: Encode the copyright information or identifier into a binary sequence.

[0173] 3. Watermark embedding: According to the positions and descriptors of the SIFT key points, select an appropriate embedding strategy and embed the encoded watermark information into the image. This can be achieved by fine-tuning the positions, directions or descriptors of the key points to ensure that the embedded watermark is invariant to scale, rotation and partial perspective changes.

[0174] 4. Watermark Detection and Extraction: When watermark detection or extraction is required, the SIFT algorithm is used again to process the image that may contain the watermark, and key points and descriptors are extracted. By comparing the SIFT descriptors of the original image and the potentially tampered image, the position of the watermark is determined. The watermark information is extracted from these positions and decoded to obtain the original copyright information or identifier.

[0175] 5. Verification and Tracing: Once the watermark is successfully extracted, its content can be verified to confirm the source of the image, audio, or video. If illegal copying or distribution is detected, the extracted watermark information can be used to trace and identify the source.

[0176] By using the SIFT method, the robustness of the watermark is ensured, enabling it to remain stable under various attacks and transformations (such as scaling, rotation, and cropping). Although the SIFT-based watermarking method has certain advantages, it also has the following disadvantages:

[0177] 1. Computational Complexity: The SIFT algorithm itself is relatively computationally complex, especially for large images or videos.

[0178] 2. Capacity Limitation: Since SIFT is a key-point-based method, the number of key points available for embedding the watermark may be limited, thus restricting the size and complexity of the watermark.

[0179] 3. Possible Distortion: Although the SIFT method aims to make the watermark stable against various transformations, excessive embedding may lead to visible distortion or a reduction in image quality.

[0180] 4. Adversarial Attacks: Although SIFT watermarks are robust against common image processing operations (such as cropping, scaling, and rotation), they may be sensitive to specially designed adversarial attacks (such as specific filtering or noise addition).

[0181] 5. Parameter Tuning: The SIFT algorithm has multiple parameters, such as the threshold for key point detection, the size of the descriptor, etc. These parameters may need to be adjusted according to the specific application and content to obtain the best watermarking effect.

[0182] The protection scope of the present invention is not limited to the above embodiments. Without departing from the spirit and scope of the inventive concept, changes and advantages that can be conceived by those skilled in the art are included in the present invention, and the appended claims define the scope of protection.

Claims

1. A digital image watermark registration and positioning method based on the straight-line Hough transform, characterized in that, The registration and positioning method includes two parts: watermark addition and watermark extraction; The watermark addition includes the following steps: Step 1: Read the information to be encoded, obtain the 256 - base data information according to the information to be encoded, process the 256 - base data information, and generate a binary watermark after encoding; Step 2: Pre - process the original image to be embedded with the watermark, calculate the texture information and edge features of the pre - processed image, generate an adaptive embedding strength matrix, extract the BGR three - channel information of the image, splice the generated basic unit watermark blocks, and embed them into the image, and reconstruct the channels of the image to obtain the image after embedding the information; The watermark extraction includes the following steps: Step I: Use an imaging recording device to obtain the image with the watermark to be extracted, and pre - process the image; Step II: Calculate the texture information and edge features of the image pre - processed in Step I to generate an adaptive embedding strength matrix, and extract the yellow - blue channel information of the image; Step III: Use the adaptive embedding strength matrix in Step II to inversely extract the spliced watermark information in the yellow - blue channel; Step IV: Use the Hough line detection algorithm to extract the positions of the lines in the spliced watermark information, and use the clustering algorithm to divide the lines into two groups that are as perpendicular as possible, calculate the coordinate positions of the intersection points of the two groups of lines, and save them according to the line grouping; Step V: Topologically expand outward from each intersection point, locate the positions of the four vertices of the upper - left, lower - left, upper - right, and lower - right corners of each block as a group, and perform perspective transformation correction on each four - vertex group to intercept the basic unit watermark block.

2. The registration and positioning method according to claim 1, characterized in that, In Step 1, the 256 - base data information includes the size of the basic unit watermark block, the encoded data information, and the error - correction information; Perform binary conversion on the 256 - base data, and generate binary mask information of the same length according to the length of the converted data; According to the size of the basic unit watermark block, a corresponding mask needs to be generated, and then the generated mask is used to add random offsets to the encoded data information and error - correction information to reduce the regularity of the data; Traverse the mask, perform exclusive - OR operation with the data after binary conversion to generate binary watermark information and the corresponding basic unit watermark block.

3. The registration and positioning method according to claim 1, characterized in that, In Step 2, the pre - processing of the original image includes adjusting the image size, adjusting the gray - scale value, and filtering; adjusting the image size means cropping the picture to retain the central part of the original image; adjusting the gray - scale value means adjusting according to the average value of all pixels in the image; the filtering process means making the extracted features obvious through filtering methods including bilateral filtering, low - pass filtering, high - pass filtering, and Wiener filtering; The generation process of the adaptive embedding strength matrix is as follows: Design a weighting function in the spatial domain of the image, and embed the watermark information into the areas of the image that are not easily recognized with higher weights; divide the image into three regions: flat region, strong - edge region, and texture region, and the judgment criteria for each region are as follows: where T = Aver Edge (I) + 2 * Std Edge (I); Among them, Aver(i,j), Std(i,j), and Edge(i,j) are the local average value, standard deviation, and edge detection value on x(i,j), respectively; x(i,j) represents the pixel value of the image; for edge detection, the Prewitt operator is used, and Aver Edge (i) and Std Edge (i) are the average value and standard deviation of the edge detection value Edge(i,j); The weighting function is as follows: Among them, Among them, α is the minimum embedding strength; WF(*) is a weighting function for dark and bright regions, and the watermark information is embedded in the dark and bright regions with a higher weight through the weighting function. The result generated from the preprocessed image according to the weighting function is the adaptive embedding strength matrix.

4. The registration and positioning method according to claim 1, characterized in that, The BGR three-channel information includes the blue channel, the green channel, and the red channel; the blue channel contains the information of the blue component of each pixel in the image; the green channel contains the information of the green component of each pixel in the image; the red channel contains the information of the red component of each pixel in the image. When the basic unit watermark blocks are tiled and spliced, the pixel interval between every two basic unit watermark blocks is 1 pixel value, which is used to embed a straight line for Hough line detection and positioning. The original yellow and blue channel information I of the image is obtained by weighting the BGR three channels B , the watermark information is w, the original image information is i, and the embedding strength matrix is the adaptive embedding strength matrix ρ, then I B ' = (1 - ρ) * I B + ρ * w; where, I B ' represents the yellow and blue channel information of the image after embedding the information, I B represents the yellow and blue channel information of the original image, and ρ represents the embedding strength, i.e., the weight; obtain the yellow and blue channel information of the image after embedding the watermark Fuse the weighted I B 'channel information with the original B, G, and R channel information to generate an RGB color image, and integrate the BGR three-channel information after embedding to obtain the image with the embedded information.

5. The registration and positioning method according to claim 1, characterized in that, In step I, the preprocessing of the watermarked image obtained by the imaging recording device refers to adjusting the image size, adjusting the gray value, and filtering. Adjusting the image size means cropping the picture and retaining the central part of the original image; adjusting the gray value means adjusting according to the average value of all pixels in the image; filtering means making the extracted features obvious through filtering methods including bilateral filtering, low-pass filtering, high-pass filtering, and Wiener filtering; bilateral filtering is used to remove salt-and-pepper noise in the image, low-pass filtering is used to smooth the image and remove high-frequency noise, and high-pass filtering is used to enhance the edges and textures of the image. Wiener filtering is used to minimize the estimation error caused by noise introduction, estimate the desired signal based on the mean square error of the original image, and is used for noise reduction and signal recovery, thereby improving the quality of the signal or image.

6. The registration and positioning method according to claim 1, characterized in that, In step III, the inverse extraction means calculating the difference between the original image and the preprocessed image, and combining it with the adaptive embedding strength matrix to obtain the embedded watermark information, that is, extracting the watermark information by dividing the difference by the adaptive embedding strength matrix.

7. The registration and positioning method according to claim 1, characterized in that, In step IV, the Hough line detection algorithm performs threshold segmentation and binarization on the watermark image after inverse extraction, sets the threshold to the pixel average value of the watermark image after inverse extraction + 20, performs Hough line detection on the binarized image, and obtains the data of the straight line. The clustering algorithms include k-means, K-Medoids, DBSCAN, Mean Shift, Agglomerative Hierarchical Clustering, which divide the straight lines into multiple groups.

8. The registration and positioning method according to claim 1, characterized in that In step V, after obtaining the intersection points of the lines, each intersection point belongs to two intersecting and approximately perpendicular lines. Starting from the current intersection point, find the second intersection point along the first line where it lies, and at the same time, find the third intersection point along the second line where it lies. Calculate the difference between the relative distance x1 between the second intersection point and the first intersection point and the relative distance x2 between the third intersection point and the first intersection point. If the difference is less than 0.1*(x1 + x2), then it is considered that these three intersection points are the three vertices of the same basic unit watermark block; otherwise, continue to find the next intersection point along the line where it lies until three points that meet the requirements are found. The fourth point is determined based on whether it lies on the same line as the second intersection point, and at the same time lies on the same line as the third intersection point, and the difference between the distance x3 from the fourth point to the second intersection point and the distance x4 from the fourth point to the third intersection point is less than 0.1*(x3 + x4). When the four vertices of a basic unit watermark block are determined, perform a perspective transformation on the parallelogram surrounded by the four vertices to transform it into a standard basic unit watermark block, and obtain the square watermark block extraction information.

9. A registration and positioning system for implementing the registration and positioning method according to any one of claims 1-8, the system comprising an encoding module, an embedding module, and an extraction module; The encoding module includes a data reading module, a random mask generation module, and a random offset module; the embedding module includes a preprocessing module, an adaptive embedding strength matrix module, and a weighting module; the extraction module includes a preprocessing module, an adaptive embedding strength matrix module, an inverse extraction module, a line detection module, and a registration and positioning module.

10. Application of the watermark registration and positioning method according to any one of claims 1-8, or the registration and positioning system according to claim 9 in copyright protection, content authentication and verification, broadcast monitoring, and fingerprint recognition.

Images