A method and device for extracting watermark of three-dimensional model

Abstract

The present application discloses a method and device for extracting watermarks from a three-dimensional model, the method comprising: acquiring multiple images of a three-dimensional model taken from multiple perspectives; for each image, selecting a potential watermark embedding area according to the grayscale gradient variation characteristics of the image in the model texture coordinate space; performing multi-scale analysis on the potential watermark embedding area to obtain the high-frequency texture characteristics of the potential watermark embedding area; performing frequency domain transformation on the potential watermark embedding area to analyze the high-frequency frequency domain characteristics of the potential watermark embedding area based on frequency domain data and high-frequency texture characteristics; fusing the high-frequency texture characteristics and high-frequency frequency domain characteristics of the potential watermark embedding area of ​​each image; performing statistical analysis on the two fused characteristics to determine the final watermark embedding area; mapping the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain a watermark extraction result.

Description

Technical Field

[0001] The present application relates to the technical field of watermark extraction, and in particular to a method and device for extracting watermarks from a three-dimensional model. Background Art

[0002] With the development of 3D digital technology, 3D models have been widely used in industrial design, cultural heritage protection and virtual reality. In order to protect the intellectual property rights of 3D digital assets, digital watermark technology has gradually become an effective means of protection. Digital watermark embeds hidden information in the texture or geometric characteristics of the 3D model, thereby storing copyright information without significantly affecting the appearance of the model. When verification is required later, the watermark in the 3D model needs to be extracted and then verified.

[0003] At present, the watermark extraction method in 3D models mainly relies on 2D image processing and frequency domain analysis. Specifically, it is mainly through texture mapping, that is, the surface of the 3D model is unfolded into a 2D texture image. Then the data in the 2D texture image is converted into the frequency domain, and then the high-frequency vector in the 2D texture image is extracted, so that the watermark can be determined by the distribution of the high-frequency vector.

[0004] However, due to the complexity and nonlinearity of the texture mapping of the 3D model, there are also shooting angles, texture occlusions, and mapping distortions, which can easily lead to the loss of texture information, so the single mapped image cannot guarantee the accuracy of the global texture of the model. In addition, it only simply analyzes the characteristics of the frequency domain without effective verification, so the existing method cannot effectively guarantee the accuracy of the extracted watermark. Summary of the invention

[0005] Based on the above-mentioned deficiencies of the prior art, the present application provides a method and device for extracting watermarks of three-dimensional models to solve the problem that the prior art cannot guarantee accurate extraction of watermarks of three-dimensional models.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] The first aspect of the present application provides a watermark extraction method for a three-dimensional model, comprising:

[0008] Acquire multiple target images of a target three-dimensional model shot from multiple perspectives;

[0009] For each of the target images, according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space, a potential watermark embedding area of ​​the target image is screened out;

[0010] Performing a multi-scale high-frequency texture characteristic analysis on the latent watermark embedding area to obtain the high-frequency texture characteristic of the latent watermark embedding area;

[0011] Performing frequency domain transformation on the latent watermark embedding region, and analyzing high-frequency frequency domain characteristics of the latent watermark embedding region based on the frequency domain data of the latent watermark embedding region and the high-frequency texture characteristics;

[0012] The high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the latent watermark embedding regions of each of the target images are respectively fused to obtain fused texture characteristics and fused frequency domain characteristics;

[0013] Determining a final watermark embedding area by statistically analyzing the fused texture characteristics and the fused frequency domain characteristics;

[0014] The fused frequency domain characteristics of the final watermark embedding area are mapped back to the model texture coordinate space to obtain a watermark extraction result of the target three-dimensional model.

[0015] Optionally, in the above-mentioned watermark extraction method for a three-dimensional model, after acquiring a plurality of target images of a target three-dimensional model shot from multiple perspectives, the method further comprises:

[0016] Using histogram equalization technology to perform contrast enhancement on each of the target image data to obtain enhanced target images;

[0017] Performing gamma correction processing on each of the enhanced target images to adjust the grayscale dynamic range of the target image;

[0018] Correcting the geometric distortion of each of the adjusted target images by homography transformation;

[0019] Adopting an adaptive median filtering algorithm to filter out noise from each of the corrected target images.

[0020] Optionally, in the above-mentioned watermark extraction method for a three-dimensional model, before screening out the potential watermark embedding area of ​​each target image according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space, the method further comprises:

[0021] For each of the target images, extract feature points in the target image;

[0022] Calculating characteristic description information of each of the feature points; wherein the characteristic description information of the feature point is information characterizing the local texture of the feature point;

[0023] Based on the texture mapping parameters of the target three-dimensional model and the camera calibration parameters when shooting the target image, a mapping model between each of the feature points and the model texture coordinate space is established;

[0024] Optimizing the mapping model by minimizing a total error function;

[0025] Calculating the local consistency of each of the feature points in the optimized mapping model;

[0026] Optimizing the feature points whose local consistency does not meet the preset conditions.

[0027] Optionally, in the above-mentioned watermark extraction method for a three-dimensional model, for each of the target images, according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space, screening out the potential watermark embedding area of ​​the target image comprises:

[0028] For each of the target images, determining a target texture region of the target image in the model texture coordinate space;

[0029] Extracting the grayscale gradient variation characteristics of each pixel point in the target texture area of ​​the target image;

[0030] Based on the grayscale gradient variation characteristics of each pixel point in the target texture region of the target image, the potential watermark embedding region of the target image is screened out through a region growing criterion.

[0031] Optionally, in the above-mentioned watermark extraction method for a three-dimensional model, the performing a multi-scale high-frequency texture characteristic analysis on the latent watermark embedding area to obtain the high-frequency texture characteristic of the latent watermark embedding area includes:

[0032] Performing multi-scale characteristic analysis on the latent watermark embedding area to obtain characteristics at multiple scales;

[0033] Calculate the difference between the scale characteristics of each two adjacent layers to obtain the initial high-frequency texture characteristics of multiple layers;

[0034] The grayscale gradient change characteristics of the potential watermark embedding area are combined with the initial texture high-frequency information of each layer to obtain an optimized high-frequency texture characteristic enhancement map.

[0035] Optionally, in the above-mentioned watermark extraction method for a three-dimensional model, after combining the grayscale gradient change characteristics of the potential watermark embedding area with the initial texture high-frequency information of each layer to obtain an optimized high-frequency texture characteristic enhancement map, the method further includes:

[0036] Boundary extraction and confirmation are performed on the high-frequency texture characteristic enhancement map, and regional consistency verification is performed.

[0037] Optionally, in the above-mentioned watermark extraction method for a three-dimensional model, the step of analyzing the high-frequency frequency domain characteristics of the latent watermark embedding area based on the frequency domain data of the latent watermark embedding area and the high-frequency texture characteristics includes:

[0038] Extracting high-frequency frequency domain data from the frequency domain data of the potential watermark embedding area;

[0039] The high-frequency frequency domain data is weighted by using the high-frequency texture characteristic enhancement map to obtain the high-frequency energy of the potential watermark embedding area;

[0040] Performing discrete wavelet transform on the frequency domain data of the potential watermark embedding area, decomposing the frequency domain data of the potential watermark embedding area into multi-level frequency domain components; wherein the frequency domain components include low frequency components, horizontal high frequency components, vertical high frequency components and diagonal high frequency components;

[0041] Calculating multi-level energy ratios using the high-frequency texture characteristic enhancement map and the frequency domain components of each level;

[0042] The high-frequency texture characteristic enhancement map, the high-frequency energy and the energy ratios of each level are combined to obtain a total characteristic component.

[0043] Optionally, in the above-mentioned watermark extraction method for a three-dimensional model, the high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the potential watermark embedding areas of each of the target images are fused to obtain fused texture characteristics and fused frequency domain characteristics, including:

[0044] According to the weights corresponding to the target attributes of each of the target images, the high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the potential watermark embedding areas of each of the target images are fused respectively to obtain fused texture characteristics and fused frequency domain characteristics; wherein the target attributes include at least shooting angle, resolution and confidence parameters.

[0045] Optionally, in the above-mentioned watermark extraction method for three-dimensional models, mapping the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain the watermark extraction result of the target three-dimensional model includes:

[0046] Mapping the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain a distribution map of the final watermark embedding area;

[0047] Acquire the position information of each feature point in the distribution map of the final watermark embedding area; wherein the position information includes texture coordinates and physical coordinates;

[0048] Calculate the confidence of each feature point according to the total feature component and the maximum feature component of each feature point in the distribution diagram of the final watermark embedding area;

[0049] Using the location information and confidence level of each feature point and the distribution map of the final watermark embedding area, a watermark extraction report is generated and output.

[0050] The second aspect of the present application provides a watermark extraction device for a three-dimensional model, comprising:

[0051] An image acquisition unit, used to acquire multiple target images of a target three-dimensional model shot from multiple perspectives;

[0052] A primary screening unit, for screening out a potential watermark embedding region of each target image according to the grayscale gradient variation characteristics of the points of the target image in the model texture coordinate space;

[0053] A texture characteristic analysis unit, used for performing multi-scale high-frequency texture characteristic analysis on the latent watermark embedding area to obtain the high-frequency texture characteristics of the latent watermark embedding area;

[0054] A frequency domain transform unit, used for performing frequency domain transform on the latent watermark embedding area;

[0055] A frequency domain characteristic analysis unit, configured to analyze the high frequency domain characteristics of the latent watermark embedding region based on the frequency domain data of the latent watermark embedding region and the high frequency texture characteristics;

[0056] An image characteristic fusion unit, used to fuse the high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the latent watermark embedding regions of each of the target images, respectively, to obtain fused texture characteristics and fused frequency domain characteristics;

[0057] A characteristic verification unit, used for determining a final watermark embedding area by statistically analyzing the fused texture characteristic and the fused frequency domain characteristic;

[0058] The result generating unit is used to map the fused frequency domain characteristics back to the model texture coordinate space to obtain the watermark extraction result of the target three-dimensional model.

[0059] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the device further comprises:

[0060] An enhancement unit, used to perform contrast enhancement on each of the target image data using a histogram equalization technique to obtain enhanced target images;

[0061] A dynamic adjustment unit, used for performing gamma correction processing on each of the enhanced target images to adjust the grayscale dynamic range of the target image;

[0062] A correction unit, used for correcting the geometric distortion of each of the adjusted target images through homography transformation;

[0063] The filtering unit is used to filter out noise from each of the corrected target images using an adaptive median filtering algorithm.

[0064] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the device further comprises:

[0065] A feature point extraction unit, used for extracting feature points in each target image respectively;

[0066] A description information calculation unit, used to calculate the characteristic description information of each of the feature points; wherein the characteristic description information of the feature point is information characterizing the local texture of the feature point;

[0067] A mapping model building unit, used to build a mapping model between each of the feature points and the model texture coordinate space based on the texture mapping parameters of the target three-dimensional model and the camera calibration parameters when shooting the target image;

[0068] An optimization unit, configured to optimize the mapping model by minimizing a total error function;

[0069] A consistency calculation unit, used for calculating the local consistency of each of the feature points in the optimized mapping model;

[0070] The optimization unit is used to optimize each of the feature points whose local consistency does not meet the preset conditions.

[0071] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the primary screening unit includes:

[0072] A region determination unit, used for determining, for each of the target images, a target texture region of the target image in the model texture coordinate space;

[0073] A gradient characteristic extraction unit, used to extract the grayscale gradient change characteristics of each pixel point in the target texture area of ​​the target image;

[0074] The characteristic screening unit is used to screen out the potential watermark embedding area of ​​the target image through a region growing criterion based on the gray gradient change characteristics of each pixel point in the target texture area of ​​the target image.

[0075] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the texture characteristic analysis unit includes:

[0076] A multi-scale analysis unit, used for performing multi-scale characteristic analysis on the latent watermark embedding area to obtain characteristics of multiple scales;

[0077] A texture characteristic calculation unit is used to calculate the difference between the scale characteristics of each two adjacent layers to obtain the initial high-frequency texture characteristics of multiple layers;

[0078] The texture characteristic combining unit is used to combine the grayscale gradient change characteristics of the potential watermark embedding area with the initial texture high-frequency information of each layer to obtain an optimized high-frequency texture characteristic enhancement map.

[0079] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the device further comprises:

[0080] The region verification unit is used to extract and confirm the boundary of the high-frequency texture characteristic enhancement map, and to perform region consistency verification.

[0081] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the frequency domain characteristic analysis unit includes:

[0082] A frequency domain data extraction unit, used to extract high-frequency frequency domain data from the frequency domain data of the potential watermark embedding area;

[0083] A frequency domain energy calculation unit, used to weight the high-frequency frequency domain data using the high-frequency texture characteristic enhancement map to obtain the high-frequency energy of the potential watermark embedding area;

[0084] A decomposition unit, used for performing discrete wavelet transform on the frequency domain data of the potential watermark embedding area, and decomposing the frequency domain data of the potential watermark embedding area into multi-level frequency domain components; wherein the frequency domain components include low frequency components, horizontal high frequency components, vertical high frequency components and diagonal high frequency components;

[0085] An energy ratio calculation unit, used to calculate the energy ratio of multiple levels by using the high-frequency texture characteristic enhancement map and the frequency domain components of each level;

[0086] The total characteristic component calculation unit is used to combine the high-frequency texture characteristic enhancement map, the high-frequency energy and the energy ratio of each level to obtain the total characteristic component.

[0087] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the image characteristic fusion unit includes:

[0088] The image characteristic fusion subunit is used to fuse the high-frequency texture characteristics and the high-frequency frequency domain characteristics of each potential watermark embedding area of ​​each target image according to the weight corresponding to the target attribute of each target image, so as to obtain fused texture characteristics and fused frequency domain characteristics; wherein the target attributes include at least shooting angle, resolution and confidence parameters.

[0089] Optionally, in the above-mentioned watermark extraction device for a three-dimensional model, the result generating unit includes:

[0090] A reverse mapping unit, used for mapping the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain a distribution map of the final watermark embedding area;

[0091] A coordinate acquisition unit, used to acquire the position information of each feature point in the distribution map of the final watermark embedding area; wherein the position information includes texture coordinates and physical coordinates;

[0092] A confidence calculation unit, configured to calculate the confidence of each of the feature points according to the total characteristic component and the maximum characteristic component of each of the feature points in the distribution diagram of the final watermark embedding area;

[0093] The report generating unit is used to generate and output a watermark extraction report by using the position information and confidence level of each of the feature points and the distribution map of the final watermark embedding area.

[0094] The present application provides a watermark extraction method for a three-dimensional model, which obtains multiple target images of a target three-dimensional model taken from multiple perspectives, so that the texture of the three-dimensional model can be accurately reflected through images from multiple perspectives, and more comprehensive and accurate characteristics can be provided. In addition, the characteristics of each image can be verified to ensure the accuracy of the characteristics, thereby improving the accuracy of watermark extraction. Then, for each target image, according to the grayscale gradient change characteristics of the pixel points of the target image in the model texture coordinate space, the potential watermark embedding area of ​​the target image is screened out, and the high-frequency texture characteristics of the potential watermark embedding area are analyzed at multiple scales to obtain the high-frequency texture characteristics of the potential watermark embedding area, thereby extracting accurate texture characteristics through multi-scale analysis. Then, the potential watermark embedding area is transformed in the frequency domain, and based on the frequency domain data and high-frequency texture characteristics of the potential watermark embedding area, the high-frequency frequency domain characteristics of the potential watermark embedding area are analyzed, so that not only the texture characteristic information is combined with the frequency domain characteristic information for analysis, but also the watermark can be more accurately located. The high-frequency texture characteristics and high-frequency frequency domain characteristics of each potential watermark embedding area of ​​each target image are fused to obtain fused texture characteristics and fused frequency domain characteristics, so as to comprehensively consider the characteristics of each image. Then, the final watermark embedding area is determined by statistically analyzing the fused texture characteristics and fused frequency domain characteristics, so as to verify the accuracy of the analyzed characteristics and obtain the accurate final watermark embedding area. Finally, the fused frequency domain characteristics are mapped back to the model texture coordinate space to obtain the watermark extraction result of the target three-dimensional model, thereby realizing a method for accurately extracting watermarks from three-dimensional models. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0096] Figure 1 A flow chart of a watermark extraction method for a three-dimensional model provided in an embodiment of the present application;

[0097] Figure 2 A flowchart of a target image preprocessing method provided in an embodiment of the present application;

[0098] Figure 3 A flowchart of a method for mapping a target image into a model texture coordinate space provided in an embodiment of the present application;

[0099] Figure 4 A flowchart of a method for screening out potential watermark embedding areas of a target image provided by an embodiment of the present application;

[0100] Figure 5 A flowchart of a method for analyzing texture characteristics provided in an embodiment of the present application;

[0101] Figure 6 A flowchart of a method for analyzing frequency domain characteristics provided in an embodiment of the present application;

[0102] Figure 7 A flow chart of a method for generating a watermark extraction result of a target three-dimensional model provided in an embodiment of the present application;

[0103] Figure 8 A schematic diagram of the architecture of a watermark extraction device for a three-dimensional model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0104] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0105] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0106] The present application embodiment provides a method for extracting watermarks from a three-dimensional model. Figure 1 As shown, the following steps are included:

[0107] S101, obtaining multiple target images of a target three-dimensional model shot from multiple perspectives.

[0108] The target three-dimensional model refers to the three-dimensional model that currently needs to perform watermark extraction.

[0109] In order to accurately restore the texture of the 3D model and obtain accurate features, in the embodiment of the present application, the target 3D model is photographed from multiple perspectives to obtain images of the target 3D model from multiple perspectives, thereby obtaining the full amount of texture through the images of the target 3D model from multiple perspectives, and then accurately restoring the texture of the 3D model, thereby improving the accuracy of watermark extraction. Moreover, the texture features of each image can also verify and complement each other, thereby improving the accuracy of the watermark finally extracted.

[0110] Optionally, the shooting angle communication satisfies the coverage of the area where the watermark is embedded. Specifically, the area may be shot at 0 to 180 degrees, thereby avoiding the loss of key texture details due to occlusion or projection distortion.

[0111] Optionally, in order to improve the quality of the target image and thus improve the accuracy of watermark extraction, in another embodiment of the present application, after executing step S101, the target image may be further preprocessed. Figure 2 As shown, a target image preprocessing method provided in an embodiment of the present application includes:

[0112] S201 , using a histogram equalization technique to perform contrast enhancement on each target image data to obtain enhanced target images.

[0113] Specifically, the contrast of the target image data is enhanced by using the histogram equalization technology, and the pixel value of the enhanced image data can be obtained as follows:

[0114]

[0115] Among them, I(x,y) is the gray value of the original target image; I min and I max are the minimum and maximum grayscale values ​​on the target image.

[0116] S202: Perform gamma correction processing on each enhanced target image to adjust the grayscale dynamic range of the target image.

[0117] Specifically, the data of the target image is adjusted by the correction coefficient, and the pixel value of the adjusted image data is obtained as follows:

[0118]

[0119] in, is a correction factor, which can be adjusted according to the characteristics of the image acquisition device. When , it is used to enhance the dark area of ​​the image, while When , it is used to correct the details of two areas of the image.

[0120] S203 , correcting the geometric distortion of each adjusted target image through homography transformation.

[0121] Changes in shooting angles may cause geometric distortion of images. In particular, when the texture and shape of the target object are projected onto the image plane at different viewing angles, the straight lines or planes in the image may appear curved or irregular. Therefore, by performing geometric correction on the image based on the homography transformation, these distorted images are adjusted back to a uniform viewing angle, so that the image is consistent with the texture mapping of the three-dimensional model.

[0122] S204: Adopt an adaptive median filtering algorithm to remove noise from each corrected target image.

[0123] Specifically, the pixel values ​​of the target image can be filtered according to a filter window of a certain size. If the pixel value is within the grayscale range of the filter window, it remains unchanged. If it exceeds the grayscale range, it can be replaced by the median of the pixel value in the window.

[0124] It should also be noted that in order to accurately analyze the texture characteristics of each target image and facilitate the subsequent fusion of the characteristics of each target image, after executing step S101 and before executing step S102, each target image needs to be mapped to the model texture coordinate space.

[0125] Specifically, a mapping relationship between the captured target image and the 3D model texture coordinates is established according to texture mapping parameters when the target 3D model is generated, and a projection change matrix is ​​specifically used to uniformly map each target image to a model texture coordinate space.

[0126] Optionally, in another embodiment of the present application, each target image is mapped to a specific implementation method in the model texture coordinate space, such as Figure 3 As shown, the following steps are included:

[0127] S301 : For each target image, extract feature points in the target image.

[0128] Optionally, multiple feature points p can be extracted from the target image i =[x i ,y i , 1] T , in order to facilitate subsequent processing; where x i ,y i is the horizontal and vertical coordinate values ​​of the feature point. Specifically, the feature point detection algorithm can be used to identify the key feature points in the texture area of ​​the target image.

[0129] S302: Calculate characteristic description information of each feature point.

[0130] The characteristic description information of the feature point is information that characterizes the local texture of the feature point, which may be composed of information in multiple dimensions.

[0131] S303: Based on the texture mapping parameters of the target three-dimensional model and the camera calibration parameters when shooting the target image, a mapping model between each feature point and the model texture coordinate space is established.

[0132] Specifically, based on the texture mapping parameters of the target 3D model and the camera calibration parameters when shooting the target image, the projection change matrix can be obtained. Then, through the projection change matrix, the data of the target image can be mapped to the model texture coordinate space, that is, a mapping model between each feature point and the model texture coordinate space is established. Therefore, for any point, the mapping relationship can be expressed as:

[0133]

[0134] Among them, H enh is the projection transformation matrix, D(p i ) are the coordinates of the feature points.

[0135] Optionally, the coordinates of the feature points after distortion correction can be used for mapping, that is, D(p i ) is the coordinate of the feature point after distortion correction. So if the distortion coefficients are k1, k2, p1 and p2, the coordinate of the feature point after distortion correction can be expressed as:

[0136]

[0137] Among them, x i and i is the coordinate of feature point i before correction; x c and c is the camera principal point coordinate; r=(x i -x c ) 2 +(y i -y c ) 2 .

[0138] S304: Optimize the mapping model by minimizing the total error function.

[0139] In order to ensure the quality after mapping, further optimization is required in the embodiment of the present application. First, the mapping relationship can be optimized by minimizing the total error function E as follows:

[0140]

[0141] Where N is the number of feature points; α is the weight coefficient for balancing reprojection error and geometric consistency, and “G(.)” is the geometric feature function, which extracts the curvature or edge direction geometric properties at the corresponding texture coordinates. The optimization process is aimed at the enhanced projection transformation matrix H enh Parameters are used.

[0142] S305: Calculate the local consistency of each feature point in the optimized mapping model.

[0143] In order to use the geometric topological consistency of the texture of the target 3D model for verification, the local consistency of each feature point is calculated:

[0144]

[0145] Among them, N(i) is the feature point p i The domain set, w ij is the weight, u i and u j Represents the coordinates of the i-th and j-th feature points after being mapped to the model texture coordinate space.

[0146] S306: Optimize each feature point whose local consistency does not meet the preset conditions.

[0147] Specifically, each feature point whose local consistency does not meet the preset conditions can be proposed to avoid the noise interference of these abnormal points on the mapping accuracy, so as to obtain an accurate mapping result and accurately reflect the texture of the target three-dimensional model.

[0148] S102 , for each target image, according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space, screen out the potential watermark embedding area of ​​the target image.

[0149] Due to the location where the embedded watermark exists, the degree of change in the grayscale value of the image will be relatively large. Therefore, in an embodiment of the present application, the grayscale value of each pixel point of the target image mapped to the model texture coordinate space is used to calculate the grayscale gradient change characteristics. Then, based on the grayscale gradient change characteristics, an image segmentation algorithm is used to screen out areas in the target image where watermarks may be embedded.

[0150] Optionally, in another embodiment of the present application, a specific implementation of step S102 is as follows: Figure 4 As shown, the following steps are included:

[0151] S401 . Determine, for each target image, a target texture region of the target image in a model texture coordinate space.

[0152] In order to speed up the processing, in the embodiment of the present application, the area where the watermark is most likely to exist will be first determined as the target texture area, or if it is the area where the consistent watermark is located, the area can be directly determined as the target texture area.

[0153] S402, extracting the grayscale gradient variation characteristics of each pixel point in the target texture area of ​​the target image.

[0154] Specifically, the grayscale gradient change is calculated based on the grayscale value of each point on the target image to obtain the gradient vector , and then the gradient assignment is obtained by approximate calculation through the Sobel operator , thus obtaining the gray gradient change characteristics of each point.

[0155] S403: Based on the grayscale gradient variation characteristics of each pixel point in the target texture region of the target image, a potential watermark embedding region of the target image is screened out by a region growing criterion.

[0156] Specifically, the pixels that meet the region growth criterion are selected, and the region where these pixels are located is the potential watermark embedding region. The region growth criterion is defined as:

[0157]

[0158] Among them, I seed is the gray value of a selected seed point; I(u i ) is the gray value of the i-th pixel; T and Assign thresholds for grayscale difference and gradient respectively.

[0159] S103, performing multi-scale high-frequency texture characteristic analysis on the latent watermark embedding area to obtain the high-frequency texture characteristics of the latent watermark embedding area.

[0160] It should be noted that since the area where the watermark is located is mainly manifested in high-frequency features, in order to obtain accurate texture characteristics and improve the accuracy of the final watermark extraction, in the implementation of this application, the high-frequency texture characteristics of the latent watermark embedding area are analyzed at multiple scales, thereby separating the background changes and embedded information in the texture of the target image, and obtaining the high-frequency texture characteristics of the latent watermark embedding area. The area where these features are located is a more accurate watermark embedding area, that is, through texture characteristic analysis, the latent watermark embedding area can be narrowed to a more accurate range.

[0161] Optionally, in another embodiment of the present application, a specific implementation of step S103 is as follows: Figure 5 As shown, the following steps are included:

[0162] S501, performing multi-scale characteristic analysis on the potential watermark embedding area to obtain characteristics of multiple scales.

[0163] Optionally, a multi-scale characteristic analysis is performed on the potential watermark embedding area, which can be specifically represented by a multi-scale Gaussian pyramid:

[0164]

[0165] Among them, “*” represents the convolution operation, and is the scale parameter of the k-th level pyramid.

[0166] S502: Calculate the difference between the scale characteristics of every two adjacent layers to obtain the initial high-frequency texture characteristics of multiple layers.

[0167] Specifically, by calculating the high-frequency features H at different scales k (u i ), and preliminarily extract the high-frequency texture characteristics related to watermark embedding:

[0168]

[0169] S503: Combine the grayscale gradient change characteristics of the potential watermark embedding area with the initial texture high-frequency information of each layer to obtain an optimized high-frequency texture characteristic enhancement map.

[0170] Specifically, after initially obtaining the high-frequency texture characteristics, the grayscale gradient change and the high-frequency texture characteristics are combined to further optimize the selected potential watermark embedding area and further narrow it, which can be specifically expressed as:

[0171]

[0172] Among them, w k is the weight of high-frequency texture features at different scales.

[0173] It should be noted that the optimization process is only to further remove the low-confidence area, that is, to narrow the potential watermark embedding area to make it the final watermark embedding area.

[0174] Optionally, in another embodiment of the present application, after executing step S503, the following may be further included:

[0175] The boundary of the high-frequency texture feature enhancement map is extracted and confirmed, and the regional consistency is verified.

[0176] Specifically, the boundary of the high-frequency texture feature enhancement map is extracted and confirmed, and the boundary points are detected, and then the regional consistency verification is performed.

[0177] S104: Perform frequency domain transformation on the potential watermark embedding area.

[0178] In order to perform watermark analysis from the frequency, in the embodiment of the present application, the frequency domain transformation is performed on the potential watermark embedding area, so that the frequency domain data of the potential watermark embedding area can be obtained.

[0179] Optionally, the grayscale signal of the potential watermark embedding area in the texture coordinate space may be converted from the space domain to the frequency domain by discrete preselection transformation, thereby obtaining the frequency domain data of the potential watermark embedding area.

[0180] Specifically, for each grayscale signal in the potential watermark embedding area in the texture coordinate space, its two-dimensional frequency domain transform, namely, DCT transform is:

[0181]

[0182] Among them, C (u, v) is the frequency coefficient matrix, that is, the frequency domain data after transformation; (u, v) is the frequency domain coordinate; and is the scaling factor; M and N represent the number of rows and columns of pixels in the region, respectively.

[0183] S105: Analyze the high-frequency frequency domain characteristics of the potential watermark embedding area based on the frequency domain data and high-frequency texture characteristics of the potential watermark embedding area.

[0184] It should be noted that the area where the watermark is located also shows high frequency in the frequency domain, so it is necessary to classify the high-frequency frequency domain characteristics. Since the high-frequency texture characteristics reflect its high-frequency characteristics in the texture, it can be used as a reference to analyze the high-frequency frequency domain characteristics of the potential watermark embedding area. In addition, the high-frequency texture characteristics can be combined to make the analyzed high-frequency frequency domain characteristics more accurate.

[0185] Optionally, in another embodiment of the present application, a specific implementation of step S105 is as follows: Figure 6 As shown, the following steps are included:

[0186] S601. Extract high-frequency frequency domain data from the frequency domain data in the potential watermark embedding area.

[0187] Specifically, the high-frequency components of the frequency domain data can be analyzed to filter out the high-frequency frequency domain data, thereby obtaining the area where the high-frequency frequency domain data is located, and then the specific location where the watermark is embedded can be further determined.

[0188] S602: weighting the high-frequency frequency domain data using the high-frequency texture characteristic enhancement map to obtain high-frequency energy in the potential watermark embedding area.

[0189] Specifically, in the embodiment of the present application, the energy in the frequency domain is used as the feature of the frequency domain to reflect the embedded position of the watermark through the distribution of energy. The high-frequency texture characteristic enhancement map also reflects the position of the watermark embedding, so it needs to be considered comprehensively. Therefore, the high-frequency texture characteristic enhancement map is combined with the high-frequency frequency domain data to calculate the high-frequency energy of the potential watermark embedding area.

[0190] Specifically, high frequency capability is defined as:

[0191]

[0192] Among them, and u t and v t is the frequency domain threshold, which combines the high-frequency texture characteristics to enhance the image E(u i )Weighted high frequency capability determination.

[0193] S603: Perform discrete wavelet transform on the frequency domain data of the potential watermark embedding area, and decompose the frequency domain data of the potential watermark embedding area into multi-level frequency domain components.

[0194] The frequency domain components include low frequency components, horizontal high frequency components, vertical high frequency components and diagonal high frequency components.

[0195] Specifically, the frequency domain data of the watermark embedding area is decomposed into low-frequency components LL k , horizontal high frequency component LH k , vertical high frequency component HL k and the diagonal high frequency component HH k , which can be specifically expressed as:

[0196]

[0197] S604, using the high-frequency texture characteristic enhancement map and the frequency domain components of each level to calculate the energy ratio of multiple levels.

[0198] In order to represent the distribution of the embedded signal in the high-frequency component, the multi-level energy ratio is calculated based on the high-frequency texture feature enhancement map, specifically:

[0199]

[0200] S605 , combining the high-frequency texture feature enhancement map, the high-frequency energy, and the energy ratios of each level to obtain a total feature component.

[0201] In order to comprehensively consider the characteristics of each part analyzed and finally extract the watermark, it is necessary to combine the analyzed high-frequency texture characteristic enhancement map, high-frequency energy and frequency domain components of each level to obtain a total characteristic component. It can be expressed as:

[0202]

[0203] in, , and is the weight coefficient, which can be optimized and adjusted based on the grayscale, frequency domain and high-frequency gradient characteristics of the embedded signal.

[0204] S106, respectively fuse the high-frequency texture characteristics and high-frequency frequency domain characteristics of each potential watermark embedding area of ​​each target image to obtain a fused texture characteristic and a fused frequency domain characteristic.

[0205] After the above analysis, the characteristics of each potential watermark embedding area in each target image are obtained. Therefore, in order to comprehensively consider the characteristics of each potential watermark embedding area in each target image and accurately analyze the position of the final watermark, the high-frequency texture characteristics and high-frequency frequency domain characteristics of each potential watermark embedding area in each target image are fused respectively to obtain fused texture characteristics and fused frequency domain characteristics.

[0206] Optionally, features belonging to the same region can be weighted according to their confidence levels, and features from different regions can be combined to obtain fused texture characteristics and fused frequency domain characteristics, that is, to obtain a distribution map of fused texture characteristics and a distribution map of fused frequency domain characteristics.

[0207] Optionally, in another embodiment of the present application, the specific implementation of step S106 includes:

[0208] According to the weights corresponding to the target attributes of each target image, the high-frequency texture characteristics and high-frequency frequency domain characteristics of each potential watermark embedding area of ​​each target image are fused respectively to obtain fused texture characteristics and fused frequency domain characteristics.

[0209] The target attributes include at least shooting angle, resolution and confidence parameters.

[0210] S107, determining the final watermark embedding area by performing statistical analysis on the fused texture characteristics and the fused frequency domain characteristics.

[0211] It should be noted that both the fused texture characteristics and the fused frequency domain characteristics can reflect the location of the watermark, so if both the fused texture characteristics and the fused frequency domain characteristics reflect that a certain location is the watermark location, that is, the distribution of the fused texture characteristics and the fused frequency domain characteristics at a certain location is consistent, then it means that it is the location of the watermark. Therefore, the fused texture characteristics are used to verify whether the distribution area of ​​the fused frequency domain characteristics accurately reflects the watermark embedding area. Specifically, the fused texture characteristics and the fused frequency domain characteristics can be counted, and then verified based on the statistical results, the final watermark embedding area can be determined, that is, the area with the features analyzed is the area where the watermark is located, so the next step can be carried out.

[0212] Specifically, the frequency domain energy and texture change characteristics of the watermark embedding area can be enhanced based on the fused high-frequency texture characteristic map and the fused total characteristic component, and the consistency between the two can be verified. The area where the two are consistent is the watermark embedding area.

[0213] S108, mapping the fused frequency domain characteristics back to the model texture coordinate space to obtain a watermark extraction result of the target three-dimensional model.

[0214] The fused frequency domain characteristics combine the texture characteristics and the frequency domain characteristics, so they can accurately reflect the watermark position, and through the previous step, it is verified that its distribution position is the watermark position. However, the fused frequency domain characteristics are frequency domain characteristics, which cannot intuitively reflect the position of the watermark, so it is necessary to map them back to the model texture coordinate space to obtain the distribution map of the characteristics of the final watermark embedding area, and this characteristic distribution map obviously presents the watermark in the target three-dimensional model. Optionally, different colors can be used for rendering according to the size of the eigenvalues, so that the watermark in the target three-dimensional model can be more intuitively presented, so that the watermark can be verified. Therefore, based on the distribution map of the final watermark embedding area, the watermark extraction result of the target three-dimensional model containing detailed information can be generated and output.

[0215] Optionally, in another embodiment of the present application, a specific implementation of step S108 is as follows: Figure 7 As shown, including:

[0216] S701, mapping the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain a distribution map of the final watermark embedding area.

[0217] S702: Obtain the position information of each feature point in the distribution map of the final watermark embedding area.

[0218] The position information includes texture coordinates and physical coordinates.

[0219] S703: Calculate the confidence of each feature point according to the total feature component and the maximum feature component of each feature point in the distribution diagram of the final watermark embedding area.

[0220] Among them, the maximum characteristic component refers to the maximum value of the total characteristic components of each feature point. Therefore, the confidence of each feature point can be expressed as:

[0221]

[0222] Among them, T c (u i ) is the total characteristic component of the feature point. max(T c ) is the maximum characteristic component.

[0223] S704: Generate and output a watermark extraction report using the location information, confidence level, and distribution map of the final watermark embedding area of ​​each feature point.

[0224] The embodiment of the present application provides a watermark extraction method for a three-dimensional model, which obtains multiple target images of a target three-dimensional model taken from multiple perspectives, so that the texture of the three-dimensional model can be accurately reflected through images from multiple perspectives, and more comprehensive and accurate characteristics can be provided. And the characteristics of each image can be verified to ensure the accuracy of the characteristics, thereby improving the accuracy of watermark extraction. Then, for each target image, according to the grayscale gradient change characteristics of the points of the target image in the model texture coordinate space, the potential watermark embedding area of ​​the target image is screened out, and the high-frequency texture characteristics of the potential watermark embedding area are analyzed at multiple scales to obtain the high-frequency texture characteristics of the potential watermark embedding area, thereby extracting accurate texture characteristics through multi-scale analysis. Then, the potential watermark embedding area is transformed in the frequency domain, and based on the frequency domain data and high-frequency texture characteristics of the potential watermark embedding area, the high-frequency frequency domain characteristics of the potential watermark embedding area are analyzed, so that not only the texture characteristics are combined with the frequency domain characteristics for analysis, but also the watermark can be more accurately located. The high-frequency texture characteristics and high-frequency frequency domain characteristics of each potential watermark embedding area of ​​each target image are fused respectively to obtain fused texture characteristics and fused frequency domain characteristics, so that the characteristics of each image can be comprehensively considered. Next, the final watermark embedding area is determined by statistically analyzing the fused texture characteristics and fused frequency domain characteristics, so that the accuracy of the analyzed features can be verified and the accurate final watermark embedding area can be obtained. Finally, the fused frequency domain characteristics are mapped back to the model texture coordinate space to obtain the watermark extraction result of the target 3D model, thus realizing a method for accurately extracting watermarks from 3D models.

[0225] Another embodiment of the present application provides a watermark extraction device for a three-dimensional model, such as Figure 8 As shown, including:

[0226] The image acquisition unit 801 is used to acquire multiple target images of a target three-dimensional model shot from multiple perspectives.

[0227] The primary screening unit 802 is used to screen out the potential watermark embedding area of ​​each target image according to the grayscale gradient variation characteristics of the points of the target image in the model texture coordinate space.

[0228] The texture characteristic analysis unit 803 is used to perform multi-scale high-frequency texture characteristic analysis on the potential watermark embedding area to obtain the high-frequency texture characteristics of the potential watermark embedding area.

[0229] The frequency domain transform unit 804 is used to perform frequency domain transform on the potential watermark embedding area.

[0230] The frequency domain characteristic analysis unit 805 is used to analyze the high-frequency frequency domain characteristics of the potential watermark embedding area based on the frequency domain data and high-frequency texture characteristics of the potential watermark embedding area.

[0231] The image characteristic fusion unit 806 is used to fuse the high-frequency texture characteristics and high-frequency frequency domain characteristics of each potential watermark embedding area of ​​each target image to obtain a fused texture characteristic and a fused frequency domain characteristic.

[0232] The characteristic verification unit 807 is used to determine the final watermark embedding area by performing statistical analysis on the fused texture characteristics and the fused frequency domain characteristics.

[0233] The result generating unit 808 is used to map the fused frequency domain characteristics back to the model texture coordinate space to obtain the watermark extraction result of the target three-dimensional model.

[0234] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the device further includes:

[0235] The enhancement unit is used to perform contrast enhancement on each target image data by using the histogram equalization technology to obtain each enhanced target image.

[0236] The dynamic adjustment unit is used to perform gamma correction processing on each enhanced target image to adjust the grayscale dynamic range of the target image.

[0237] The correction unit is used to correct the geometric distortion of each adjusted target image through homography transformation.

[0238] The filtering unit is used to filter out noise from each corrected target image by using an adaptive median filtering algorithm.

[0239] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the device further includes:

[0240] The feature point extraction unit is used to extract the feature points in each target image respectively.

[0241] The description information calculation unit is used to calculate the characteristic description information of each feature point, wherein the characteristic description information of the feature point is information representing the local texture of the feature point.

[0242] The mapping model building unit is used to build a mapping model between each feature point and the model texture coordinate space based on the texture mapping parameters of the target three-dimensional model and the camera calibration parameters when shooting the target image.

[0243] The optimization unit is used to optimize the mapping model by minimizing the total error function.

[0244] The consistency calculation unit is used to calculate the local consistency of each feature point in the optimized mapping model.

[0245] The optimization unit is used to optimize each feature point whose local consistency does not meet the preset conditions.

[0246] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the primary screening unit includes:

[0247] The region determination unit is used to determine the target texture region of the target image in the model texture coordinate space for each target image.

[0248] The gradient characteristic extraction unit is used to extract the grayscale gradient change characteristics of each pixel point in the target texture area of ​​the target image.

[0249] The characteristic screening unit is used to screen out the potential watermark embedding area of ​​the target image through the region growing criterion based on the gray gradient change characteristics of each pixel point in the target texture area of ​​the target image.

[0250] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the texture characteristic analysis unit includes:

[0251] The multi-scale analysis unit is used to perform multi-scale characteristic analysis on the potential watermark embedding area to obtain characteristics of multiple scales.

[0252] The texture characteristic calculation unit is used to calculate the difference between the scale characteristics of each two adjacent layers to obtain the initial high-frequency texture characteristics of multiple layers.

[0253] The texture characteristic combining unit is used to combine the grayscale gradient variation characteristics of the potential watermark embedding area with the initial texture high-frequency information of each layer to obtain an optimized high-frequency texture characteristic enhancement map.

[0254] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the device further includes:

[0255] The region verification unit is used to extract and confirm the boundaries of the high-frequency texture feature enhancement map, as well as to verify the region consistency.

[0256] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the frequency domain characteristic analysis unit includes:

[0257] The frequency domain data extraction unit is used to extract high-frequency frequency domain data from the frequency domain data in the potential watermark embedding area.

[0258] The frequency domain energy calculation unit is used to weight the high-frequency frequency domain data using the high-frequency texture characteristic enhancement map to obtain the high-frequency energy of the potential watermark embedding area.

[0259] The decomposition unit is used to perform discrete wavelet transform on the frequency domain data of the potential watermark embedding area, and decompose the frequency domain data of the potential watermark embedding area into multi-level frequency domain components, wherein the frequency domain components include low frequency components, horizontal high frequency components, vertical high frequency components and diagonal high frequency components.

[0260] The energy ratio calculation unit is used to calculate the energy ratio of multiple levels by utilizing the high-frequency texture characteristic enhancement map and the frequency domain components of each level.

[0261] The total characteristic component calculation unit is used to combine the high-frequency texture characteristic enhancement map, the high-frequency energy and the energy ratio of each level to obtain the total characteristic component.

[0262] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the image characteristic fusion unit includes:

[0263] The image characteristic fusion subunit is used to fuse the high-frequency texture characteristics and high-frequency frequency domain characteristics of each potential watermark embedding area of ​​each target image according to the weight corresponding to the target attribute of each target image, so as to obtain fused texture characteristics and fused frequency domain characteristics. Among them, the target attributes include at least shooting angle, resolution and confidence parameters.

[0264] Optionally, in the watermark extraction device for a three-dimensional model provided in another embodiment of the present application, the result generating unit includes:

[0265] The reverse mapping unit is used to map the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain the distribution map of the final watermark embedding area.

[0266] The coordinate acquisition unit is used to obtain the position information of each feature point in the distribution map of the final watermark embedding area.

[0267] The position information includes texture coordinates and physical coordinates.

[0268] The confidence calculation unit is used to calculate the confidence of each feature point according to the total characteristic component and the maximum characteristic component of each feature point in the distribution diagram of the final watermark embedding area.

[0269] The report generation unit is used to generate and output a watermark extraction report using the location information, confidence level and distribution map of the final watermark embedding area of ​​each feature point.

[0270] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the implementation process of the corresponding steps in the above method embodiments, and will not be repeated here.

[0271] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0272] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A watermark extraction method for a three-dimensional model, characterized in that: include: Acquire multiple target images of a target three-dimensional model shot from multiple perspectives; For each of the target images, according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space, a potential watermark embedding area of ​​the target image is screened out; Performing a multi-scale high-frequency texture characteristic analysis on the latent watermark embedding area to obtain the high-frequency texture characteristic of the latent watermark embedding area; Performing frequency domain transformation on the latent watermark embedding region, and analyzing high-frequency frequency domain characteristics of the latent watermark embedding region based on the frequency domain data of the latent watermark embedding region and the high-frequency texture characteristics; The high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the latent watermark embedding regions of each of the target images are respectively fused to obtain fused texture characteristics and fused frequency domain characteristics; Determining a final watermark embedding area by statistically analyzing the fused texture characteristics and the fused frequency domain characteristics; The fused frequency domain characteristics of the final watermark embedding area are mapped back to the model texture coordinate space to obtain a watermark extraction result of the target three-dimensional model.

2. The method according to claim 1, characterized in that: After acquiring a plurality of target images of the target three-dimensional model shot from multiple perspectives, the method further includes: Using histogram equalization technology to perform contrast enhancement on each of the target image data to obtain enhanced target images; Performing gamma correction processing on each of the enhanced target images to adjust the grayscale dynamic range of the target image; Correcting the geometric distortion of each of the adjusted target images by homography transformation; Adopting an adaptive median filtering algorithm to filter out noise from each of the corrected target images.

3. The method according to claim 1, characterized in that Before filtering out the potential watermark embedding area of ​​each target image according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space, the method further includes: For each of the target images, extract feature points in the target image; Calculating characteristic description information of each of the feature points; wherein the characteristic description information of the feature point is information characterizing the local texture of the feature point; Based on the texture mapping parameters of the target three-dimensional model and the camera calibration parameters when shooting the target image, a mapping model between each of the feature points and the model texture coordinate space is established; Optimizing the mapping model by minimizing a total error function; Calculating the local consistency of each of the feature points in the optimized mapping model; Optimizing the feature points whose local consistency does not meet the preset conditions.

4. The method according to claim 1, characterized in that: The method of selecting a potential watermark embedding area of ​​each target image according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space comprises: For each of the target images, determining a target texture region of the target image in the model texture coordinate space; Extracting the grayscale gradient variation characteristics of each pixel point in the target texture area of ​​the target image; Based on the grayscale gradient variation characteristics of each pixel point in the target texture region of the target image, the potential watermark embedding region of the target image is screened out through a region growing criterion.

5. The method according to claim 1, characterized in that The performing multi-scale high-frequency texture characteristic analysis on the latent watermark embedding area to obtain the high-frequency texture characteristic of the latent watermark embedding area includes: Performing multi-scale characteristic analysis on the latent watermark embedding area to obtain characteristics at multiple scales; Calculate the difference between the scale characteristics of each two adjacent layers to obtain the initial high-frequency texture characteristics of multiple layers; The grayscale gradient change characteristics of the potential watermark embedding area are combined with the initial texture high-frequency information of each layer to obtain an optimized high-frequency texture characteristic enhancement map.

6. The method according to claim 5, characterized in that After combining the grayscale gradient change characteristics of the potential watermark embedding area with the initial texture high-frequency information of each layer to obtain an optimized high-frequency texture characteristic enhancement map, the method further includes: Boundary extraction and confirmation are performed on the high-frequency texture characteristic enhancement map, and regional consistency verification is performed.

7. The method according to claim 5, characterized in that The step of analyzing the high-frequency frequency domain characteristics of the potential watermark embedding area based on the frequency domain data of the potential watermark embedding area and the high-frequency texture characteristics comprises: Extracting high-frequency frequency domain data from the frequency domain data of the potential watermark embedding area; The high-frequency frequency domain data is weighted by using the high-frequency texture characteristic enhancement map to obtain the high-frequency energy of the potential watermark embedding area; Performing discrete wavelet transform on the frequency domain data of the potential watermark embedding area, decomposing the frequency domain data of the potential watermark embedding area into multi-level frequency domain components; wherein the frequency domain components include low frequency components, horizontal high frequency components, vertical high frequency components and diagonal high frequency components; Calculating multi-level energy ratios using the high-frequency texture characteristic enhancement map and the frequency domain components of each level; The high-frequency texture characteristic enhancement map, the high-frequency energy and the energy ratios of each level are combined to obtain a total characteristic component.

8. The method according to claim 1, characterized in that: The step of fusing the high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the latent watermark embedding regions of each of the target images to obtain fused texture characteristics and fused frequency domain characteristics comprises: According to the weights corresponding to the target attributes of each of the target images, the high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the potential watermark embedding areas of each of the target images are fused respectively to obtain fused texture characteristics and fused frequency domain characteristics; wherein the target attributes include at least shooting angle, resolution and confidence parameters.

9. The method according to claim 7, characterized in that: Mapping the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain the watermark extraction result of the target three-dimensional model includes: Mapping the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain a distribution map of the final watermark embedding area; Acquire the position information of each feature point in the distribution map of the final watermark embedding area; wherein the position information includes texture coordinates and physical coordinates; Calculate the confidence of each feature point according to the total feature component and the maximum feature component of each feature point in the distribution diagram of the final watermark embedding area; Using the location information and confidence level of each feature point and the distribution map of the final watermark embedding area, a watermark extraction report is generated and output.

10. A watermark extraction device for a three-dimensional model, characterized in that: include: An image acquisition unit, used to acquire multiple target images of a target three-dimensional model shot from multiple perspectives; A primary screening unit, for screening out a potential watermark embedding region of each target image according to the grayscale gradient variation characteristics of the pixel points of the target image in the model texture coordinate space; A texture characteristic analysis unit, used for performing multi-scale high-frequency texture characteristic analysis on the latent watermark embedding area to obtain the high-frequency texture characteristics of the latent watermark embedding area; A frequency domain transform unit, used for performing frequency domain transform on the latent watermark embedding area; A frequency domain characteristic analysis unit, configured to analyze the high frequency domain characteristics of the latent watermark embedding region based on the frequency domain data of the latent watermark embedding region and the high frequency texture characteristics; An image characteristic fusion unit, used to fuse the high-frequency texture characteristics and the high-frequency frequency domain characteristics of each of the latent watermark embedding regions of each of the target images, respectively, to obtain fused texture characteristics and fused frequency domain characteristics; A characteristic verification unit, used for determining a final watermark embedding area by statistically analyzing the fused texture characteristic and the fused frequency domain characteristic; The result generating unit is used to map the fused frequency domain characteristics of the final watermark embedding area back to the model texture coordinate space to obtain the watermark extraction result of the target three-dimensional model.

Images