Watermark embedding and copyright protection method based on signed distance field

Abstract

The invention discloses a watermark embedding and copyright protection method based on a signed distance field, and the method comprises the steps: obtaining a plurality of sampling points of an original signed distance field function F space, and enabling the sampling points to be distributed on different contour surfaces; adjusting the distance between the adjacent contour surfaces, embedding watermark information, and fitting a new symbol distance field function; and based on a to-be-verified model and the new fitted symbolic distance field function, watermark verification is carried out, and watermark embedding and copyright protection of the symbolic distance field are completed. According to the method, the watermark is embedded by changing the distance between the adjacent contour surfaces in the SDF space, and the deformation distance is used as watermark information for verification, so that the effective protection of the copyright of the model is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of copyright protection of neural network models, and in particular relates to a watermark embedding and copyright protection method based on a signed distance field. Background Art

[0002] With the rapid development of 3D computer graphics, virtual reality, augmented reality and digital content generation technology (AI Generated Content, AIGC), 3D geometric modeling based on signed distance field (SDF) has been widely used. SDF is a method that represents the distance of a point in 3D space by mapping it to a geometric surface. It can accurately describe complex 3D shapes and allow arbitrarily parsing and extracting 3D mesh models from SDF. Due to its continuity and high precision, SDF has become an important tool for 3D geometric representation.

[0003] At the same time, with the development of AIGC technology, especially the rise of technologies such as Implicit Neural Fields, the representation ability of SDF in complex scenes has been further improved, enabling SDF-based neural network models to generate high-precision 3D models through data-driven learning. Supported by technologies such as Neural Radiance Fields (NeRF), such models have been widely used in image generation, 3D reconstruction, virtual reality, augmented reality, and other AIGC scenarios. However, any AIGC network model enhanced with SDF has the potential to extract the original mesh, which shows that identifying and protecting the intellectual property rights of SDF in AIGC models has become a pressing issue.

[0004] Traditional copyright protection technologies for 3D models and neural network models have the following problems: Digital watermarking technology for discrete mesh models: It is usually embedded in discretized meshes, but because the mesh structure is vulnerable to re-meshing, simplification and geometric transformation, the watermark information is easily invalid and cannot cope with the disturbance and deformation of the complex geometric structure of the model. NeRF (neural radiant field) watermarking technology: For example, CopyRNeRF mainly embeds watermarks by changing color information. However, this method ignores the protection of geometric information. Attackers can bypass watermark verification by extracting geometric meshes, thereby illegally using the generated AIGC model. SDF watermarking technology based on local deformation: There are also some methods that divide the space into multiple regions through a spherical coordinate system, embed watermark information in each region, and change the distance between isosurfaces through local deformation to achieve protection of SDF and its generated meshes. However, such methods will lead to a significant increase in computational complexity in high-resolution meshes or large 3D models, especially when the model data generated in AIGC scenarios is huge, which brings huge storage and computational burdens. The limitation of these traditional methods is that they usually only protect discrete meshes or color information, but ignore the continuity and robustness of SDF in geometric representation. Although the local deformation SDF watermarking method has proposed improvements to some problems, the increase in computational complexity is still its limitation.

[0005] Therefore, the present invention proposes a watermark embedding and copyright protection method based on signed distance field. Summary of the invention

[0006] In order to solve the above technical problems, the present invention proposes a watermark embedding and copyright protection method based on signed distance field. By making slight adjustments to the distances between isosurfaces, the watermark information is embedded in a specific interval in the SDF. This method can achieve watermark embedding without affecting the geometric accuracy of the model, thereby realizing copyright protection of the SDF and neural network model.

[0007] To achieve the above object, the present invention provides a watermark embedding and copyright protection method based on a signed distance field, comprising:

[0008] Acquire a number of sampling points of the original signed distance field function F space, wherein the sampling points are distributed on different isosurfaces; adjust the distances between adjacent isosurfaces to embed watermark information, and fit a new signed distance field function;

[0009] Based on the model to be verified and the newly fitted signed distance field function, watermark verification is performed to complete watermark embedding and copyright protection of the signed distance field.

[0010] Optionally, adjusting the distance between adjacent isosurfaces to embed watermark information, and fitting a new signed distance field function includes:

[0011] Select ray direction based on the acquired sampling points;

[0012] Based on each pair of adjacent iso-surfaces, the initial distance between adjacent iso-surfaces is calculated;

[0013] Adjacent isosurface intervals are selected for watermark embedding according to the geometric structure of the model;

[0014] Deform the sampling points in the selected isosurface interval and adjust the isosurface spacing;

[0015] Record and save the isosurface spacing and corresponding watermark information after each deformation;

[0016] According to the deformed sampling points, a new signed distance field function is fitted.

[0017] Optionally, selecting the ray direction includes: the selected ray direction needs to be optimized according to characteristics of the geometric model, so that adjusting the isosurface distance in the selected ray direction has minimal impact on the overall shape of the model.

[0018] Optionally, the sampling points in the selected isosurface interval are deformed, and the isosurface interval is adjusted including:

[0019] If the watermark 0 is embedded, the sampling point moves inward along the ray direction, shortening the distance to the lower isosurface;

[0020] If watermark 1 is embedded, the sampling points move outwards, increasing the distance from the upper isosurface.

[0021] Optionally, fitting a new signed distance field function according to the deformed sampling points includes:

[0022] (x, y, z) new =(x,y,z)+δ(x,y,z),

[0023] SDF new =SDF ori ,

[0024] Among them, (x, y, z) is the original coordinate of the sampling point, δ(x, y, z) is the deformation distance of the sampling point, (x, y, z) new is the coordinate of the sampling point after deformation, SDF ori is the original conforming distance field function, SDF new is the deformed signed distance field function.

[0025] Optionally, performing watermark verification based on the model to be verified and the newly fitted signed distance field function includes:

[0026] Submitting the model to be verified and the newly fitted signed distance field function to the verification party as a basis for preliminary verification;

[0027] The verification party extracts isosurface data according to the fitted new signed distance field function to obtain the distance change between the isosurface position and the adjacent isosurfaces;

[0028] The verifier extracts the isosurface spacing according to the model to be verified and compares it with the distance change recorded when the watermark is embedded, and determines whether the change conforms to the watermark embedding rule;

[0029] The verifier locates the watermark embedding interval of the model to be verified by comparison, and confirms whether the watermark embedding interval matches the known watermark information;

[0030] Extract watermark data according to the located embedding interval, analyze the deformation direction and amplitude of the isosurface, and identify the embedded watermark information;

[0031] The verification party compares the extracted watermark information with the recorded original watermark data to determine whether the model to be verified has any infringement behavior;

[0032] After completing the watermark extraction and comparison, the verification party generates a verification report.

[0033] Optionally, determining whether the model to be verified has any infringement includes:

[0034] If the extracted watermark information does not match the recorded watermark information, the verification party determines that the model to be verified has been tampered with, the watermark has been removed, or other infringements have occurred.

[0035] Optionally, the verification report includes the SDF analysis results of the model to be verified, the watermark extraction process, the comparison results with the original watermark, and the final determination of whether there is any infringement.

[0036] Technical effects of the present invention:

[0037] (1) The present invention embeds watermarks by directly adjusting the distance between adjacent isosurfaces. This method is relatively simple and effective, avoiding the complex operations of traditional watermarking technology. Only a slight adjustment of the distance between isosurfaces is required, thereby significantly reducing the amount of calculation when embedding and decoding watermarks, and is suitable for processing larger-scale 3D data models. In addition, the method of adjusting the distance between isosurfaces makes watermark embedding more flexible, and any ray direction and different areas can be selected. By setting the size of the adjustment distance, watermarks of different capacities can be flexibly embedded.

[0038] (2) The present invention embeds watermarks by adjusting the distance between isosurfaces, and can maintain the watermark effect when the model undergoes geometric transformation (such as scaling, translation, and rotation). Due to the properties of function space watermarks, the method of the present invention is robust to classic re-gridding attacks. Compared with traditional watermarking schemes, the method of the present invention can better maintain the robustness of the watermark under geometric deformation.

[0039] (3) The present invention only makes minor adjustments to the distance between isosurfaces. The range and effect of the adjustment can be flexibly controlled by setting thresholds and conditions, and the SDF function is not changed. This approach can better control the impact of watermark embedding on the accuracy of the geometric model, especially in the current application scenario, and can maintain the geometric fidelity of the model. By adjusting the distance between isosurfaces, the overall shape change of the model is reduced, avoiding the inconsistency or excessive distortion that may be caused by local deformation, and maintaining the quality of the model.

[0040] (4) The present invention verifies the watermark by the change in the distance between the isosurfaces, and directly determines the copyright of the SDF and the neural network model according to the change in the distance, which simplifies the encoding process. It only needs to detect the distribution of isosurfaces in a specific area to determine the existence of the watermark. Compared with the traditional complex decoding and watermark verification method, this verification method is simpler and more efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0042] Figure 1 A schematic diagram of a process of a watermark embedding and copyright protection method based on a signed distance field according to an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of adjacent isosurfaces according to an embodiment of the present invention; DETAILED DESCRIPTION

[0044] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0046] like Figure 1As shown, this embodiment provides a watermark embedding and copyright protection method based on a signed distance field, including:

[0047] Acquire a number of sampling points of the original signed distance field function F space, wherein the sampling points are distributed on different isosurfaces; adjust the distances between adjacent isosurfaces to embed watermark information, and fit a new signed distance field function;

[0048] Based on the model to be verified and the fitting of a new signed distance field function, watermark verification is performed to complete the watermark embedding and copyright protection of the signed distance field.

[0049] Furthermore, adjusting the distance between adjacent isosurfaces to embed watermark information, and fitting a new signed distance field function includes:

[0050] Select ray direction based on the acquired sampling points;

[0051] Based on each pair of adjacent iso-surfaces, the initial distance between adjacent iso-surfaces is calculated;

[0052] Adjacent isosurface intervals are selected for watermark embedding according to the geometric structure of the model;

[0053] Deform the sampling points in the selected isosurface interval and adjust the isosurface spacing;

[0054] Record and save the isosurface spacing and corresponding watermark information after each deformation;

[0055] According to the deformed sampling points, a new signed distance field function is fitted.

[0056] Specifically, the acquisition of sampling points includes: selecting multiple sampling points from the signed distance field (SDF) function F space, and these points are located on different isosurfaces. The selection of sampling points should cover the key surface features of the geometric model to ensure that the geometric structure remains reasonable during the deformation process. For more complex geometric structures, the accuracy of embedded watermarks can be improved by increasing the sampling density of specific areas.

[0057] Selecting the ray direction includes: the developer selects the normal direction along the geometric surface as the ray direction for embedding the watermark. Because the normal direction changes relatively stably in the geometric shape, adjusting the distance between adjacent isosurfaces along this direction has little effect on the overall geometry, which is convenient for subsequent verification and detection. If the model has symmetry or special geometric features, the ray direction can be optimized according to the model structure.

[0058] like Figure 2As shown in the figure, calculating the distance between isosurfaces includes: for each sampling point, calculating the initial distance between it and the adjacent isosurfaces. These distances are used as the basis for subsequent deformation adjustment to ensure that the distance change is controlled within a reasonable range during the watermark embedding process and does not affect the overall performance of the model. The initial distance between each pair of adjacent isosurfaces is recorded to provide a reference for subsequent verification.

[0059] Selecting the embedding interval includes: developers select the interval between adjacent isosurfaces for watermark embedding according to the geometric structure and surface features of the model. For example, selecting the intervals [0,0.1] and [0.1,0.3] for embedding means embedding "0" or "1". The distance change of these intervals will represent part of the watermark information. By selecting different intervals and sampling point positions, the watermark embedding is ensured to be concealed while avoiding affecting the geometric structure.

[0060] Adjusting the isosurface spacing includes: embedding watermarks by fine-tuning the sampling points within the isosurface interval according to the developer's needs. Embedding watermark "0": Move the sampling points inward along the ray direction to shorten the isosurface distance within the interval. For example, adjust the isosurface distance in the [0,0.1] interval from the initial value 0.1 to 0.08 to reduce the distance in the interval. Embedding watermark "1": Move the sampling points outward along the ray direction to increase the isosurface distance in the interval. For example, adjust the isosurface distance in the [0.1,0.3] interval from 0.2 to 0.22 to expand the distance in the interval.

[0061] Recording and saving deformation information includes: each time the isosurface distance is adjusted, the deformation amplitude and specific watermark information are recorded. This deformation information will be used in subsequent verification steps to ensure that the watermark can be accurately detected. Save all deformation data, including the change in isosurface distance, ray direction, and corresponding watermark interval to ensure that the watermark embedding area can be quickly located during verification.

[0062] Fitting a new SDF function includes: generating a new SDF function G based on the adjusted sampling points and isosurface distances, which reflects the slight changes in the distances of adjacent isosurfaces in a specific interval, while the rest of the area remains unchanged. The new SDF function can still maintain the same geometric representation as the original model and will not affect the performance of the model. However, there is a slight change in the isosurface distance in a specific interval, which is used to verify the copyright of the model. The new SDF function is fitted as follows:

[0063] (x, y, z) new =(x,y,z)+δ(x,y,z) (1),

[0064] SDF new =SDF ori (2)

[0065] Among them, (x, y, z) is the original coordinate of the sampling point, δ(x, y, z) is the deformation distance of the sampling point, (x, y, z) new is the coordinate of the sampling point after deformation, SDF ori is the original conforming distance field function, SDF new is the deformed signed distance field function.

[0066] Where δ(x, y, z) is the offset between specific isosurfaces in the ray direction, which is used to embed watermark information. This offset can be positive or negative, depending on the "inward" or "outward" deformation to be achieved; Formula (2) indicates that the SDF function after embedding the watermark is the same as the original SDF function in terms of the three-dimensional geometric structure, and embedding the watermark does not affect the geometric model.

[0067] Further, based on the model to be verified and the newly fitted signed distance field function, watermark verification includes:

[0068] Submitting the model to be verified and the newly fitted signed distance field function to the verification party as a basis for preliminary verification;

[0069] The verification party extracts isosurface data according to the fitted new signed distance field function to obtain the distance change between the isosurface position and the adjacent isosurfaces;

[0070] The verifier extracts the isosurface spacing according to the model to be verified and compares it with the distance change recorded when the watermark is embedded, and determines whether the change conforms to the watermark embedding rule;

[0071] The verifier locates the watermark embedding interval of the model to be verified by comparison, and confirms whether the watermark embedding interval matches the known watermark information;

[0072] Extract watermark data according to the located embedding interval, analyze the deformation direction and amplitude of the isosurface, and identify the embedded watermark information;

[0073] The verification party compares the extracted watermark information with the recorded original watermark data to determine whether the model to be verified has any infringement behavior;

[0074] After completing the watermark extraction and comparison, the verification party generates a verification report.

[0075] Specifically, submitting the model and SDF function to be verified includes: after the model is published or distributed, if there is suspected infringement, the model owner can submit the SDF function to be verified to the verifier. The function will serve as the basis for preliminary verification.

[0076] Extracting isosurface data includes: the verifier extracts the distance data between isosurfaces from the SDF function to be verified, focuses on analyzing the specific interval where the watermark is embedded, and records the actual distance between adjacent isosurfaces. By calculating the distance change between isosurfaces, the possible location of the embedded watermark is determined.

[0077] Comparing the isosurface distance changes includes: the verifier compares the isosurface distance extracted from the model to be verified with the initial distance data recorded by the developer when embedding the watermark. By analyzing the distance changes in each interval, the verifier can determine whether these deformations conform to the deformation rules when embedding the watermark.

[0078] Locating the interval where the watermark is embedded includes: the verifier locates the specific interval where the watermark is embedded by analyzing the isosurface spacing, and confirms whether it meets the expected distance change. By comparing the original and deformed isosurface spacings, the interval where the watermark is embedded is determined.

[0079] Extracting watermark information includes: the verification party confirms the interval of watermark embedding according to the distance change data and extracts the watermark information from it. By analyzing the amplitude and direction of the isosurface deformation, the verification party can extract the embedded watermark data and verify the authenticity of the watermark embedding.

[0080] Comparing the extracted watermark with the original watermark includes the verification party comparing the extracted watermark distance change with the recorded original watermark data to confirm whether they are consistent. If they are consistent, the model passes the verification; if they are inconsistent, there may be tampering or infringement.

[0081] Determining infringement includes: If the extracted watermark distance does not match the record, the verifier can determine that the model may have watermark tampering or infringement. The verifier makes an infringement judgment based on the comparison results.

[0082] Generating a verification report includes: the verifier generates a report to record the isosurface data extracted during the verification process and the comparison results. If there is any infringement, the verification report will detail the evidence of the infringement.

[0083] Legal and copyright protection measures include: If infringement is confirmed, the verifier will provide relevant evidence to assist the model owner in taking legal action. The verification report can be used as a legal basis for copyright protection and related litigation procedures.

[0084] The present invention proposes a watermark embedding and copyright protection method based on a signed distance field. The watermark is embedded by changing the distance between adjacent isosurfaces on a ray in the SDF space, and the copyright protection of the model can be performed without affecting the SDF function. Usually, the signed distance field (SDF) represents three-dimensional geometric shapes through isosurfaces and serves as an important geometric input of the neural network model. The distance between isosurfaces is the key to the construction of geometric shapes. The isosurface dimensions in the SDF can be adjusted by control functions, and different dimensions have different effects on the geometric representation. The larger isosurface distance plays a major role in the geometric representation of the model, while the isosurface fluctuations in a small range have a reduced effect on the overall geometric accuracy and have almost no effect on the overall performance of the model. Therefore, by making a slight adjustment to the distance between the isosurfaces and embedding the watermark information into a specific interval in the SDF, the watermark embedding can be achieved without affecting the geometric accuracy of the model, and the copyright protection of the SDF and the neural network model can be achieved.

[0085] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A watermark embedding and copyright protection method based on signed distance field, characterized in that: include: Acquire a number of sampling points of the original signed distance field function F space, wherein the sampling points are distributed on different isosurfaces; adjust the distances between adjacent isosurfaces to embed watermark information, and fit a new signed distance field function; Based on the model to be verified and the newly fitted signed distance field function, watermark verification is performed to complete watermark embedding and copyright protection of the signed distance field.

2. The watermark embedding and copyright protection method based on signed distance field as claimed in claim 1, characterized in that: Adjusting the distance between adjacent isosurfaces to embed watermark information and fitting a new signed distance field function comprises: Select ray direction based on the acquired sampling points; Based on each pair of adjacent iso-surfaces, the initial distance between adjacent iso-surfaces is calculated; Adjacent isosurface intervals are selected for watermark embedding according to the geometric structure of the model; Deform the sampling points in the selected isosurface interval and adjust the isosurface spacing; Record and save the isosurface spacing and corresponding watermark information after each deformation; According to the deformed sampling points, a new signed distance field function is fitted.

3. The watermark embedding and copyright protection method based on signed distance field as claimed in claim 2, characterized in that: Selecting the ray direction includes: the selected ray direction needs to be optimized according to the characteristics of the geometric model, so that adjusting the isosurface distance in the selected ray direction has the least effect on the overall shape of the model.

4. The watermark embedding and copyright protection method based on signed distance field as claimed in claim 2, characterized in that: Deform the sampling points in the selected isosurface interval and adjust the isosurface spacing including: If the watermark 0 is embedded, the sampling point moves inward along the ray direction, shortening the distance to the lower isosurface; If watermark 1 is embedded, the sampling points move outwards, increasing the distance from the upper isosurface.

5. The watermark embedding and copyright protection method based on signed distance field as claimed in claim 2, characterized in that: According to the deformed sampling points, the new signed distance field function is fitted, including: (x,y,z) new (x,y,z)+δ(x,y,z), <h2 style=";text-align:left;direction:ltr">SDF<h2 style=";text-align:left;direction:ltr"> new <h2 style=";text-align:left;direction:ltr"> (SDF)<h2 style=";text-align:left;direction:ltr"> ori <h2 style=";text-align:left;direction:ltr"> , Among them, (x, y, z) is the original coordinate of the sampling point, δ(x, y, z) is the deformation distance of the sampling point, (x, y, z) new is the coordinate of the sampling point after deformation, SDF ori is the original conforming distance field function, SDF new is the deformed signed distance field function.

6. The watermark embedding and copyright protection method based on signed distance field as claimed in claim 1, characterized in that: Based on the model to be verified and the newly fitted signed distance field function, watermark verification includes: Submitting the model to be verified and the newly fitted signed distance field function to the verification party as a basis for preliminary verification; The verification party extracts isosurface data according to the fitted new signed distance field function to obtain the distance change between the isosurface position and the adjacent isosurfaces; The verifier extracts the isosurface spacing according to the model to be verified and compares it with the distance change recorded when the watermark is embedded, and determines whether the change conforms to the watermark embedding rule; The verifier locates the watermark embedding interval of the model to be verified by comparison, and confirms whether the watermark embedding interval matches the known watermark information; Extract watermark data according to the located embedding interval, analyze the deformation direction and amplitude of the isosurface, and identify the embedded watermark information; The verification party compares the extracted watermark information with the recorded original watermark data to determine whether the model to be verified has any infringement behavior; After completing the watermark extraction and comparison, the verification party generates a verification report.

7. The watermark embedding and copyright protection method based on signed distance field as claimed in claim 6, characterized in that: Determining whether the model to be verified has infringement includes: If the extracted watermark information does not match the recorded watermark information, the verification party determines that the model to be verified has been tampered with, the watermark has been removed, or other infringements have occurred.

8. The watermark embedding and copyright protection method based on signed distance field as claimed in claim 6, characterized in that: The verification report includes the SDF analysis results of the model to be verified, the watermark extraction process, the comparison results with the original watermark, and the final determination of whether there is any infringement.

Images