Model denoising method and electronic device

Abstract

The application provides a model denoising method and an electronic device. The method comprises the following steps: acquiring position information of each vertex in a current round of a to-be-denoised model and patch information of each triangular patch, and determining a plurality of virtual sphere radii according to the position information of each vertex; determining a volume proportion of a current vertex according to the virtual sphere radii, the patch information, and the position information of the current vertex, and determining a control operator of the current vertex according to the volume proportion corresponding to each virtual sphere radius of the current vertex and a weight coefficient; determining a guide normal according to the position information of the current vertex, the patch information, and the control operator, and updating the position information of the current vertex according to the control operator, the guide normal of the current vertex, and the position information of each vertex. The application realizes high-precision and high-robustness mesh model denoising by designing a plurality of virtual sphere radii and calculating a control operator, and is suitable for multi-resolution and multi-complexity industrial workpiece mesh processing scenes.

Description

Technical Field

[0001] This application relates to the field of model building technology, and more specifically, to a model denoising method and an electronic device. Background Technology

[0002] 3D optical scanning technology has been widely used in industrial manufacturing, reverse engineering, and precision measurement. Through scanning methods such as structured light, line laser, and binocular stereo vision, 3D mesh models of workpiece surfaces can be quickly acquired for digital reconstruction and quality inspection. However, due to the optical properties of the workpiece material, equipment system errors, environmental vibrations, and ambient light interference, 3D mesh models may contain high-frequency noise and topological distortion patches, making them unsuitable for direct high-precision reverse modeling and dimensional measurement. Therefore, how to perform mesh denoising and smoothing has become a pressing technical problem to be solved.

[0003] In existing technologies, algorithms such as Laplace smoothing, bilateral filtering, or differential geometric principal curvature estimation are commonly used to denoise the mesh.

[0004] However, for differential geometric principal curvature estimation algorithms, the differentiation operation amplifies high-frequency noise, making it difficult to adjust the feature recognition threshold parameter, failing to accurately distinguish features from noise, and exhibiting poor noise robustness. For bilateral filtering algorithms, if the curvature threshold is increased or the filtering window is widened to filter out high-frequency noise, the algorithm may misclassify minute geometric features and faint text on the workpiece itself as noise and smooth them with high intensity, resulting in over-smoothing. If the parameters are tightened, complete denoising is not possible, leading to under-smoothing. For Laplacian smoothing and differential geometric principal curvature estimation algorithms, the feature extraction scale is strongly tied to the grid resolution. Curvature values ​​drift non-linearly under different scanning resolutions, and fixed parameters cannot achieve stable denoising, lacking scale robustness. Summary of the Invention

[0005] The purpose of this application is to provide a model denoising method and electronic device to address the shortcomings of the prior art, thereby solving the problems of poor noise robustness, scale robustness, and smoothing in the prior art.

[0006] To achieve the above objectives, the technical solution adopted in this application is as follows: Firstly, this application provides a model denoising method, the method comprising: Obtain the position information of each vertex and the face information of each triangle in the model to be denoised in the current round, and determine the radii of multiple virtual spheres based on the position information of each vertex; Traverse each vertex. For the current vertex, determine the volume percentage of the current vertex under each virtual sphere radius based on the radius of each virtual sphere, the face information of multiple triangles, and the position information of the current vertex. Each virtual sphere radius indicates a virtual sphere range. The volume percentage is used to characterize the proportion of the overlap between the virtual sphere range and the model to be denoised within the virtual sphere range. The control operator for the current vertex is determined based on the volume percentage of the current vertex under each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius. Based on the position information of the current vertex, the face information of multiple triangles, and the control operator of the current vertex, the guiding normal of the current vertex is determined; The position information of the current vertex is updated based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex.

[0007] Optionally, determining the volume percentage of the current vertex corresponding to each virtual sphere radius based on the radius of each virtual sphere, the facet information of the plurality of triangular facets, and the position information of the current vertex includes: Based on the virtual sphere radius, determine the initial sampling point of the current vertex within the virtual sphere range corresponding to the virtual sphere radius, wherein the distance between the initial sampling point corresponding to the virtual sphere radius and the current vertex is less than or equal to the virtual sphere radius; Based on the face information of the multiple triangular facets and the position information of the current vertex, multiple internal sampling points are selected from multiple initial sampling points. The internal sampling points are located inside the model to be denoised. Based on the number of internal sampling points and the number of initial sampling points, the volume percentage of the current vertex corresponding to the radius of the virtual sphere is determined.

[0008] Optionally, determining the initial sampling point of the current vertex within the virtual sphere range corresponding to the virtual sphere radius, based on the virtual sphere radius, includes: Using the current vertex as the center, construct a bounding box for the current vertex based on the radius of the virtual sphere; A preset number of discrete sampling points are generated within the bounding box of the current vertex; Discrete sampling points whose distance from the current vertex is less than or equal to the radius of the virtual sphere are used as initial sampling points corresponding to the radius of the virtual sphere of the current vertex.

[0009] Optionally, the step of selecting multiple internal sampling points from multiple initial sampling points based on the facet information of the multiple triangular facets and the position information of the current vertex includes: Based on the information of each triangular facet, determine the position of the center point of each triangular facet; Using the current vertex as the topological starting point, a breadth-first search is performed on the mesh surface of the model to be denoised using a pre-constructed topology table to obtain at least one initial triangular facet. Each initial triangular facet is then added to the local manifold facet set of the current vertex, wherein the distance between the center point of each initial triangular facet and the current vertex is less than or equal to the radius of the virtual sphere. Based on the set of local manifold patches and the patch information of each triangular patch in the model to be denoised, multiple internal sampling points are selected from multiple initial sampling points.

[0010] Optionally, the step of selecting multiple internal sampling points from multiple initial sampling points based on the set of local manifold patches and the patch information of each triangular patch in the model to be denoised includes: In the set of local manifold patches, determine the target triangular patch that is closest to each initial sampling point; Based on the center point position of the target triangle, the normal of the triangle in the triangle's information, and the position information of each initial sampling point, determine the discrimination factor of each initial sampling point of the current vertex; The initial sampling point whose discrimination factor is less than or equal to the first preset value is taken as the internal sampling point of the current vertex.

[0011] Optionally, determining the control operator for the current vertex based on the volume percentage of the current vertex at each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius includes: The absolute value of the difference between the volume percentage corresponding to the radius of each virtual sphere of the current vertex and the second preset value is taken as the feature volume percentage corresponding to the radius of each virtual sphere of the current vertex. The multi-scale features of the current vertex are determined based on the proportion of each feature volume corresponding to each virtual sphere radius of the current vertex and the preset weight coefficients corresponding to each virtual sphere radius. The control operator for the current vertex is determined based on the multi-scale features of the current vertex and the preset sensitivity parameters.

[0012] Optionally, determining the guiding normal of the current vertex based on the position information of the current vertex, the face information of multiple triangles, and the control operator of the current vertex includes: Based on the position information of the current vertex and the face information of multiple triangular faces, determine multiple neighboring faces of the current vertex; The initial normal of the current vertex is determined based on the patch information of multiple neighboring patches; The feature guidance weights are determined based on the control operator, the patch normals in the patch information of multiple neighboring patches, and the initial normal of the current vertex. The guiding normal of the current vertex is determined based on the normals of multiple neighboring faces, the feature guiding weights, and the preset distance weights.

[0013] Optionally, updating the position information of the current vertex based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex includes: Based on the position information of each vertex, determine multiple neighboring vertices of the current vertex; Based on the position information of each neighboring vertex, the position information of the current vertex, and the preset cotangent weight coefficient, the same-pole components are determined; Based on the position information of each neighboring vertex, the position information of the current vertex, and the guiding normal of the current vertex, the heterogeneous components are determined; The position information of the current vertex is updated based on the preset filter step size control factor, the control operator, the same-pair component, and the different-pair component.

[0014] Optionally, determining the radii of multiple virtual spheres based on the position information of each vertex includes: Based on the position information of each vertex, determine the global average side length of the model to be denoised; The global average side length is amplified by multiple amplification factors to obtain multiple virtual sphere radii.

[0015] Secondly, this application provides a model denoising apparatus, the apparatus comprising: The acquisition module is used to acquire the position information of each vertex and the face information of each triangle in the model to be denoised in the current round, and to determine the radius of multiple virtual spheres based on the position information of each vertex. The first determining module is used to traverse each vertex. For the current vertex that has been traversed, based on the radius of each virtual sphere, the face information of the multiple triangular facets, and the position information of the current vertex, it determines the volume ratio of the current vertex under each virtual sphere radius. Each virtual sphere radius indicates a virtual sphere range. The volume ratio is used to characterize the proportion of the overlap between the virtual sphere range and the model to be denoised in the virtual sphere range. The second determining module is used to determine the control operator of the current vertex based on the volume ratio of the current vertex under each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius; The third determining module is used to determine the guiding normal of the current vertex based on the position information of the current vertex, the face information of multiple triangles, and the control operator of the current vertex. The update module is used to update the position information of the current vertex based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex.

[0016] Thirdly, this application provides an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the model denoising method described above.

[0017] Fourthly, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the model denoising method described above.

[0018] The beneficial effects of this application are as follows: It obtains the position information of each vertex and the facet information of each triangle in the model to be denoised in the current round. Based on the position information of each vertex, it determines multiple virtual sphere radii, traverses each vertex, and for the current vertex, determines the volume proportion of the current vertex under each virtual sphere radius based on the virtual sphere radius, the facet information of multiple triangles, and the position information of the current vertex. Based on the volume proportion of the current vertex under each virtual sphere radius and the weight coefficients corresponding to each virtual sphere radius, it determines the control operator of the current vertex. Based on the position information of the current vertex, the facet information of multiple triangles, and the control operator of the current vertex, it determines the guiding normal of the current vertex. Based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex, it updates the position information of the current vertex. This application avoids mesh resolution limitations and has scale robustness by designing multiple virtual sphere radii. By calculating the volume ratio based on a virtual sphere to obtain the control operators for each vertex, the unidirectional noise reduction and anisotropic feature protection are dynamically balanced. Then, the position information of the current vertex is iteratively updated according to the guiding normal and the control operators. This allows for the precise preservation of the microscopic multi-scale geometric features of the workpiece while filtering out high-frequency noise from the 3D mesh scan, achieving high-precision and robust mesh model denoising, and adapting to multi-resolution and multi-complex industrial workpiece mesh processing scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic flowchart of a model denoising method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the range of a virtual sphere provided in an embodiment of this application; Figure 3 This is a schematic diagram of a process for determining volume percentage provided in an embodiment of this application; Figure 4 This is a schematic diagram of a process for determining initial sampling points provided in an embodiment of this application; Figure 5 This is a schematic diagram of a process for filtering internal sampling points provided in an embodiment of this application; Figure 6 This is a schematic diagram of a process for filtering internal sampling points provided in an embodiment of this application; Figure 7 This is a schematic diagram of a process for determining a control operator provided in an embodiment of this application; Figure 8 This is a schematic flowchart illustrating the process of determining a guide normal according to an embodiment of this application; Figure 9 This is a schematic diagram of a process for updating location information provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a model denoising device provided in an embodiment of this application; Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0022] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0023] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0024] Existing technologies typically employ algorithms such as Laplacian smoothing, bilateral filtering, or differential geometric principal curvature estimation to denoise meshes. However, these methods suffer from poor noise robustness, oversmoothing, undersmoothing, and a lack of scale robustness.

[0025] Based on this, this application proposes a model denoising method. This method generates control operators for each vertex by calculating the volume ratio of each vertex at multiple scales. It dynamically adjusts the isotropic denoising component and the anisotropic feature-preserving smoothing component, and iteratively updates the position of each vertex in the model. This application can automatically switch smoothing strategies based on the geometric structure of the model. While effectively removing high-frequency noise, it accurately preserves sharp edges and detailed features, and has strong robustness to different noise levels and mesh densities.

[0026] Next, refer to Figure 1 The specific implementation of the model denoising method is introduced. Among them, Figure 1 This is a flowchart illustrating a model denoising method provided in an embodiment of this application.

[0027] S101. Obtain the position information of each vertex and the face information of each triangle in the model to be denoised in the current round, and determine the radii of multiple virtual spheres based on the position information of each vertex.

[0028] The denoising model is a 3D mesh model acquired by a 3D optical scanning device, such as a structured light scanning device, a line laser scanning device, or a binocular stereo vision inspection device. The denoising model may contain random high-frequency noise and topologically distorted patches, and is composed of multiple vertices and multiple triangular patches. The vertex position information includes the 3D coordinates of the vertices and the neighborhood relationships between them. The triangular patch information can be composed of the vertices, normals, and spatial position parameters of the triangular patches.

[0029] Optionally, the model to be denoised can be represented as vertex information , For the number of vertices, face information , This represents the number of triangular facets.

[0030] Specifically, the global average side length is determined based on the position information of each vertex, and multiple virtual sphere radii are determined based on the global average side length. The number of virtual sphere radii can be pre-set according to the feature complexity of the model to be denoised; models with higher feature complexity correspond to a larger number of virtual sphere radii. For example, a complex workpiece with engravings and multiple shallow grooves can be set with 3 virtual sphere radii, while a simple workpiece with a smooth surface can be set with 2 virtual sphere radii.

[0031] S102. Traverse each vertex. For the current vertex, determine the volume percentage of the current vertex under each virtual sphere radius based on the radius of each virtual sphere, the face information of multiple triangles, and the position information of the current vertex. Each virtual sphere radius indicates a virtual sphere range, and the volume percentage is used to characterize the proportion of the overlap between the virtual sphere range and the model to be denoised within the virtual sphere range.

[0032] Figure 2 This is a schematic diagram of the range of a virtual sphere provided in an embodiment of this application. For example... Figure 2 As shown, taking the model to be denoised as an example, corresponding to two virtual sphere radii R1 and R2, each virtual sphere radius corresponds to a virtual sphere range, which is a three-dimensional spherical space range.

[0033] As an optional implementation, the model interface corresponding to the radius of each virtual sphere is determined based on the surface information of each surface within the virtual sphere range, and the volume ratio corresponding to the radius of each virtual sphere is determined based on the radius of each virtual sphere and the model interface corresponding to the radius of each virtual sphere.

[0034] As another optional implementation, based on the radius of each virtual sphere, multiple virtual sphere ranges are constructed with the current vertex as the center. Multiple initial sampling points are found within each virtual sphere range. Then, based on the facet information of multiple triangular patches and the position information of the current vertex, multiple internal sampling points are selected from the multiple initial sampling points. For each virtual sphere radius of the current vertex, the sum of the volumes of all initial sampling points and the sum of the volumes of all internal sampling points corresponding to each virtual sphere radius are calculated. The quotient of the sum of the volumes of internal sampling points and the sum of the volumes of initial sampling points is taken as the volume ratio corresponding to each virtual sphere radius.

[0035] Optionally, the volume percentage can be used to quantify the planar, convex, and concave geometric features of a local region of a vertex, thereby distinguishing effective geometric features from high-frequency scanning noise. Specifically, if the model within the virtual sphere of the current vertex is planar, the volume percentage approaches 0.5; if it is a convex region with edges, the volume percentage is greater than 0.5; if it is a shallow groove or concave region, the volume percentage is less than 0.5; and the volume percentage of noise regions approaches the planar value infinitely, without significant fluctuations.

[0036] S103. Determine the control operator for the current vertex based on the volume ratio of the current vertex under each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius.

[0037] The weight coefficient corresponding to each virtual sphere radius is a pre-configured normalized fusion coefficient for each virtual sphere radius, and the sum of the weight coefficients corresponding to each virtual sphere radius is 1.

[0038] Each vertex can correspond to a control operator, which is used to dynamically adjust the isotropic smoothing strength and anisotropic feature protection strength during the model denoising process. For example, when the comprehensive eigenvalue of vertices in the planar region of the model approaches 0, the control operator approaches 1, thereby achieving global isotropic smoothing and eliminating noise. When the comprehensive eigenvalue of vertices in the model's edges and imprinted feature regions increases, the control operator approaches 0, thereby triggering anisotropic feature protection.

[0039] S104. Determine the guiding normal of the current vertex based on the position information of the current vertex, the face information of multiple triangles, and the control operator of the current vertex.

[0040] Among them, the guiding normal can be the optimal fitting normal of the vertex. The guiding normal is used to constrain the update direction of the vertex position, ensuring that the edges of the feature region of the model are not blurred and the planar region is smooth and uniform.

[0041] Since the original normal of a vertex simply fits the normals of its neighboring surfaces without constraining the features, the smoothing process can blur small features and blunt sharp edges. Therefore, a guiding normal for the current vertex is determined based on the control operator of the current vertex, thereby adapting to the normal requirements of the feature region and the planar region.

[0042] Optionally, for vertices in complex feature regions of the model, since the control operator approaches 0, the calculation of the vertex's guiding normal only retains the neighboring surface normals that are close to the vertex's initial normal and consistent with the feature direction to participate in the weighting, thereby preserving complex features. For vertices in smooth regions of the model, since the control operator approaches 1, the vertex's guiding normal refers to all neighboring surface normals to participate in the weighting, thereby achieving noise reduction and smoothing.

[0043] S105. Update the position information of the current vertex based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex.

[0044] Specifically, the position information of the current vertex can be updated based on a position iteration formula. This formula includes homotropic components, heterotropic components, and the current position term. Both homotropic and heterotropic components are adjusted by control operators, and the heterotropic component is influenced by a guiding normal, ensuring that the vertex position information updated in the current iteration adapts to multi-scale geometric features. For locally smooth regions of the model, the homotropic component has a higher proportion, effectively filtering out surface burr noise. For locally complex regions of the model, such as machined sharp edges and engraved areas, the heterotropic component has a higher proportion, smoothing only tangential noise while preserving edge contours.

[0045] Optionally, after the update is completed, the face information of each triangle and the vertex information of each vertex of the updated model to be denoised are recalculated, and it is determined whether the iteration cutoff condition is met. If yes, the updated model to be denoised is used as the target model; otherwise, steps S101 to S105 are repeated until the iteration cutoff condition is met. The iteration cutoff condition can be a preset number of iterations or a global root mean square error threshold.

[0046] In this embodiment, the position information of each vertex and the facet information of each triangle in the model to be denoised in the current round are obtained. Based on the position information of each vertex, multiple virtual sphere radii are determined. Each vertex is traversed. For the current vertex, based on the virtual sphere radii, the facet information of the multiple triangles, and the position information of the current vertex, the volume proportion of the current vertex under each virtual sphere radius is determined. Based on the volume proportion of the current vertex under each virtual sphere radius and the weight coefficients corresponding to each virtual sphere radius, the control operator of the current vertex is determined. Based on the position information of the current vertex, the facet information of the multiple triangles, and the control operator of the current vertex, the guiding normal of the current vertex is determined. Based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex, the position information of the current vertex is updated. In this embodiment, by designing multiple virtual sphere radii, the mesh resolution limitation is avoided, and scale robustness is achieved. By calculating the volume ratio based on a virtual sphere to obtain the control operators for each vertex, the unidirectional noise reduction and anisotropic feature protection are dynamically balanced. Then, the position information of the current vertex is iteratively updated according to the guiding normal and the control operators. This allows for the precise preservation of the microscopic multi-scale geometric features of the workpiece while filtering out high-frequency noise from the 3D mesh scan, achieving high-precision and robust mesh model denoising, and adapting to multi-resolution and multi-complex industrial workpiece mesh processing scenarios.

[0047] Figure 3This is a schematic diagram illustrating a process for determining volume percentage according to an embodiment of this application. For example... Figure 3 As shown, the process of determining the volume percentage of the current vertex under each virtual sphere radius in step S102 above, based on the radius of each virtual sphere, the facet information of multiple triangular facets, and the position information of the current vertex, will be explained next.

[0048] S301. Based on the virtual sphere radius, determine the initial sampling point of the current vertex within the virtual sphere range corresponding to the virtual sphere radius.

[0049] Among them, the distance between the initial sampling point corresponding to each virtual sphere radius and the current vertex is less than or equal to the virtual sphere radius.

[0050] S302. Based on the face information of multiple triangular facets and the position information of the current vertex, select multiple internal sampling points from multiple initial sampling points. The internal sampling points are located inside the model to be denoised.

[0051] Specifically, the local manifold patch retrieval is performed using the current vertex position information. Based on the patch normal and center point position of the triangular patch, each initial sampling point is spatially verified to filter out the initial sampling points located on the workpiece entity side of the model to be denoised, thereby obtaining the internal sampling points.

[0052] like Figure 2 As shown, the red circle indicates the range of the virtual sphere. Within any virtual sphere, there are multiple initial sampling points. Some of these initial sampling points are also internal sampling points. It can be seen that the internal sampling points are located inside the model to be denoised.

[0053] S303. Determine the volume percentage of the current vertex within the virtual sphere radius based on the number of internal sampling points and the number of initial sampling points.

[0054] Optionally, the quotient of the number of internal sampling points to the number of initial sampling points can be used as the volume percentage of the current vertex within the radius of the virtual sphere.

[0055] In this embodiment, the initial sampling points corresponding to the current vertex under the radius of each virtual sphere are first determined. Then, multiple internal sampling points are selected from the multiple initial sampling points based on the facet information of the triangular facets. Based on the number of internal sampling points and the number of initial sampling points, the volume ratio corresponding to the current vertex under the radius of the virtual sphere is determined, thereby avoiding the problem of amplifying high-frequency scanning noise by the differential operator and improving the robustness of feature recognition.

[0056] Figure 4 This is a schematic diagram illustrating a process for determining initial sampling points provided in an embodiment of this application. For example... Figure 4As shown, the following describes the specific process in step S301 above of determining the initial sampling point of the current vertex within the virtual sphere range corresponding to the virtual sphere radius, based on the virtual sphere radius: S401. Using the current vertex as the center, construct the bounding box of the current vertex based on the radius of the virtual sphere.

[0057] The bounding box is an axis-aligned cube with the current vertex as the center and a side length that is twice the radius of the corresponding virtual sphere.

[0058] Specifically, taking the current vertex's 3D coordinates as the center, local axis-aligned bounding boxes are constructed along the X, Y, and Z coordinate axes, with the virtual sphere radius as the positive and negative offsets. These bounding boxes completely and tightly enclose the virtual sphere space of the corresponding scale. Simultaneously, to adapt to the efficient query requirements of dense meshes and support parallel computation by the Graphics Processing Unit (GPU), a spatial acceleration structure, either an axis-aligned bounding box tree or a linear octree, is pre-constructed in the GPU memory based on the vertex coordinates of all vertices in the model. This optimizes the traditionally time-consuming global spatial distance retrieval into a range query with logarithmic time complexity.

[0059] Specifically, the bounding box is expressed as follows: (1) (1) in, Let be the bounding box corresponding to the radius of the s-th virtual sphere at the i-th vertex. , as well as The three-dimensional coordinates of the current vertex. Let be the radius of the s-th virtual sphere.

[0060] S402. Generate a preset number of discrete sampling points within the bounding box of the current vertex.

[0061] Here, discrete sampling points are sets of points generated within the bounding box according to a uniform grid or pseudo-random sequence rule, and can be represented as follows: , This represents the total number of discrete sampling points.

[0062] Specifically, within a regular bounding box space, a predetermined fixed number of spatial discrete sampling points are generated in batches through uniform grid division or pseudo-random sequence.

[0063] S403. Take the discrete sampling points whose distance from the current vertex is less than or equal to the radius of the virtual sphere as an initial sampling point corresponding to the radius of the virtual sphere of the current vertex.

[0064] Specifically, the bounding box space is larger than the virtual sphere space, and there are a large number of invalid sampling points outside the sphere within the bounding box. By filtering using a Euclidean distance threshold, invalid sampling points outside the virtual sphere are removed, and only valid sampling points inside the sphere are retained.

[0065] Specifically, the calculation process for determining the initial sampling points is shown in the following formula (2): (2) in, This provides the location information for the m-th initial sampling point. This provides the position information for the i-th vertex. Let be the radius of the s-th virtual sphere.

[0066] In this embodiment, by constructing the bounding box of the current vertex and finding multiple initial sampling points corresponding to the virtual sphere radius of the current vertex within the bounding box, the efficiency of sampling point generation and retrieval is improved.

[0067] Next, refer to Figure 5 The process of selecting multiple internal sampling points from multiple initial sampling points in step S302 above, based on the facet information of multiple triangular faces and the position information of the current vertex, is described below. Figure 5 This is a schematic diagram of a process for filtering internal sampling points provided in an embodiment of this application.

[0068] S501. Determine the center point position of each triangular facet based on the facet information of each triangular facet.

[0069] Specifically, the center point position of each triangular facet is calculated by averaging the three-dimensional coordinates of the three vertices of the triangular facet.

[0070] As an alternative implementation, the center point of the triangular facet can be determined from the facet information itself.

[0071] S502. Using the current vertex as the topological starting point, perform a breadth-first search on the mesh surface of the model to be denoised using a pre-built topology table to obtain at least one initial triangular facet. Add each initial triangular facet to the local manifold facet set of the current vertex, wherein the distance between the center point of each initial triangular facet and the current vertex is less than or equal to the radius of the virtual sphere.

[0072] The topology table is a pre-built mesh adjacency list or semi-linked list topology structure used to record the topological associations between mesh vertices and triangular faces.

[0073] Specifically, starting with the current vertex, a pre-built topology table of mesh adjacencies is invoked to perform a breadth-first search that expands layer by layer on the mesh surface. Connected triangles are continuously traversed until the physical distance between the center of a triangle and the current vertex exceeds the radius of the current virtual sphere. All triangles that meet the distance requirement and are topologically connected are used as initial triangles, and the set of all obtained initial triangles is used as the set of local manifold faces of the current vertex.

[0074] Specifically, the process of determining the initial triangular facets can be referred to the following formula (3): (3) in, Let be the coordinates of the center point of the k-th triangular facet.

[0075] S503. Based on the set of local manifold patches and the patch information of each triangular patch in the model to be denoised, select multiple internal sampling points from multiple initial sampling points.

[0076] Specifically, based on the set of local manifold patches, the initial sampling points within the range of all virtual spheres are checked one by one according to the patch normals and center point positions of each triangular patch. Sampling points located in the overlapping area between the model and the virtual sphere are selected as internal sampling points. In this embodiment, by using a pre-built topology table to perform a breadth-first search on the mesh surface of the model to be denoised, the set of local manifold patches is determined, and then internal sampling points are selected from the initial sampling points. This effectively isolates the interference from back patches and invalid topology patches caused by the thin-walled structure of the workpiece, and greatly improves the accuracy of internal sampling point selection.

[0077] Figure 6 This is a schematic diagram illustrating a process for filtering internal sampling points according to an embodiment of this application. The following refers to... Figure 6 The process of selecting multiple internal sampling points from multiple initial sampling points in step S503 above, based on the set of local manifold patches and the patch information of each triangular patch in the model to be denoised, is described.

[0078] S601. In the set of local manifold patches, determine the target triangular patch that is closest to each initial sampling point.

[0079] Specifically, since the local geometric features of the model are determined by the nearest triangle facets, distant facets have no geometric constraint on the current initial sampling point. Therefore, each initial sampling point is traversed, and for each traversed initial sampling point, a set of local manifold facets corresponding to the current initial sampling point is determined. The spatial distance between the current initial sampling point and each triangle facet is calculated one by one, and the triangle facet with the smallest spatial distance is selected as the target triangle facet corresponding to the current initial sampling point.

[0080] S602. Based on the position of the center point of the target triangle, the normal of the triangle in the triangle information, and the position information of each initial sampling point, determine the discrimination factor of each initial sampling point of the current vertex.

[0081] Specifically, the coordinates of the center point of the target triangular facet and the facet normal are obtained, the position vector of the sampling point relative to the center of the facet is constructed, and the dot product operation of the position vector and the facet normal is performed to obtain the discrimination factor corresponding to each initial sampling point.

[0082] Alternatively, the discriminant factor can be obtained based on the following formula (4): (4) in, The discriminant factor for the m-th initial sampling point. The location of the center point of the target triangular facet. It is the normal of the target triangular facet. This represents the location information of the m-th initial sampling point.

[0083] Specifically, if the discrimination factor is less than or equal to 0, it indicates that the initial sampling point is geometrically located inside the fluid surface where the current vertex is located, that is, inside the model. If the discrimination factor is greater than 0, it indicates that the initial sampling point is geometrically located outside the fluid surface where the current vertex is located, that is, outside the model.

[0084] S603. The initial sampling point whose discrimination factor is less than or equal to the first preset value is taken as the internal sampling point of the current vertex.

[0085] Optionally, the first preset value can be 0.

[0086] Specifically, the calculated discrimination factor of each initial sampling point is compared with 0, and all initial sampling points with a discrimination factor less than or equal to 0 are selected as internal sampling points.

[0087] In this embodiment, the discriminant factors of each initial sampling point of the current vertex are used to filter internal sampling points, thereby improving the robustness of feature recognition and accurately filtering sampling points within the model.

[0088] Figure 7 This is a schematic flowchart illustrating the determination of a control operator according to an embodiment of this application. Next, refer to... Figure 7 The step of determining the control operator of the current vertex based on the volume ratio of the current vertex under each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius in step S103 above will be introduced.

[0089] S701. The absolute value of the difference between the volume percentage corresponding to the radius of each virtual sphere of the current vertex and the second preset value is used as the feature volume percentage corresponding to the radius of each virtual sphere of the current vertex.

[0090] Among them, the feature volume ratio is the absolute value of the deviation of the volume ratio under the radius of each virtual sphere from the ideal plane reference. It is used to characterize the significance of the geometric features of the local region of the current vertex. The larger the deviation, the more significant the geometric concavity and convexity features of the current vertex under the virtual sphere range.

[0091] Optionally, the second preset value can be 0.5.

[0092] Specifically, the process of determining the feature volume ratio can be shown in the following formula (5): (5) in, The characteristic volume percentage of the i-th vertex in the radius of the s-th virtual sphere. Let be the volume percentage of the i-th vertex in the radius of the s-th virtual sphere.

[0093] S702. Determine the multi-scale features of the current vertex based on the proportion of each feature volume corresponding to the radius of each virtual sphere of the current vertex and the preset weight coefficients corresponding to the radius of each virtual sphere.

[0094] Among them, the multi-scale feature is the global comprehensive feature quantity of the vertex, which is used to characterize the local multi-scale geometric concavity and convexity properties of the vertex.

[0095] Specifically, multi-scale features can be determined by referring to the following formula (6). The following formula (6) is illustrated with the number of virtual sphere radii being 2.

[0096] (6) in, For the multi-scale feature of the i-th vertex. The weighting coefficient for the radius of the first virtual sphere. The weighting coefficient for the radius of the second virtual sphere. Let be the proportion of the characteristic volume of the i-th vertex within the radius of the first virtual sphere. This represents the proportion of the feature volume of the i-th vertex within the radius of the second virtual sphere. The sum of the weight coefficients corresponding to the radii of each virtual sphere is 1. For example, It is 0.4. It is 0.6.

[0097] S703. Determine the control operator for the current vertex based on the multi-scale characteristics of the current vertex and the preset sensitivity parameters.

[0098] Among them, the preset sensitivity parameter can be used to adjust the steepness of the nonlinear mapping of the control operator, so as to achieve adaptive perception accuracy adjustment of feature strength.

[0099] For example, in the flat curved surface area of ​​the workpiece, the multi-scale features approach 0 and the control operator approaches 1, so the region uniform noise reduction is performed. In the complex areas such as workpiece markings, sharp corners and shallow grooves, the multi-scale features approach 1 and the control operator approaches 0, so the algorithm weakens smoothing and prioritizes the protection of geometric features.

[0100] Specifically, the control operator of the current vertex can be determined by referring to the following formula (7): (7) in, For the control operator of the i-th vertex, To preset the sensitivity parameters, Let be the multi-scale feature of the i-th vertex.

[0101] In this embodiment, by determining the multi-scale features of the current vertex, and then determining the control operator of the current vertex based on the multi-scale features of the current vertex and the preset sensitivity parameters, the problem of incomplete recognition of single-scale features is solved by taking into account both micro-fine textures and macro-contour features.

[0102] Figure 8 This is a schematic flowchart illustrating the process of determining a guide normal according to an embodiment of this application. (The following refers to...) Figure 8 The process of determining the guiding normal of the current vertex in step S104 above, based on the position information of the current vertex, the face information of multiple triangles, and the control operator of the current vertex, will be described.

[0103] S801. Based on the position information of the current vertex and the face information of multiple triangles, determine the multiple neighborhood faces of the current vertex.

[0104] Specifically, based on the neighborhood relationships between vertices in the current vertex's position information and the face information of each triangular facet, multiple neighborhood faces corresponding to the current vertex are retrieved.

[0105] S802. Determine the initial normal of the current vertex based on the face information of multiple neighboring faces.

[0106] Specifically, the initial normal of the current vertex before feature optimization can be calculated using the neighborhood normal average fitting algorithm.

[0107] S803. Determine the feature guidance weights based on the control operator, the surface normals in the surface information of multiple neighboring surfaces, and the initial normal of the current vertex.

[0108] Optionally, feature guidance weights can be determined based on the normal similarity kernel function, according to the control operator, the normals in the normal information of multiple neighboring faces, and the initial normal of the current vertex.

[0109] Specifically, the feature guidance weights of the current vertex can be determined based on the following formula (8): (8) in, It is a feature-guided weight. For the control operator of the i-th vertex, The normal similarity kernel function is... Let be the normal of the j-th neighboring surface. Let be the initial normal to the i-th vertex.

[0110] Specifically, when the control operator approaches 0, the feature-guided weights are dominated by normal similarity; when the control operator approaches 1, the similarity constraint becomes ineffective, and the weights tend to 1.

[0111] S804. Determine the guiding normal of the current vertex based on the normals of multiple neighboring faces, the guiding weights of features, and the preset distance weights.

[0112] Specifically, for the normals of each neighboring face, preset distance weights and feature guidance weights are simultaneously superimposed, and a weighted summation operation is performed. The summed normal vector is then normalized to obtain the guidance normal of the current vertex.

[0113] Specifically, the guiding normal of the current vertex can be determined based on the following formula (9): (9) in, It is the guiding normal of the i-th vertex. Let i be the set of neighborhood faces of the i-th vertex. For distance weights.

[0114] In this embodiment, the guiding normal of the current vertex is determined based on the normals of multiple neighboring facets, feature guiding weights, and preset distance weights. Only valid normals in the same direction are retained in the feature region to avoid blurring of edges and subtle features by normal averaging. Normals are fully fused in the planar region to ensure a smooth surface and noise reduction effect.

[0115] Figure 9 This is a schematic diagram of a process for updating location information provided in an embodiment of this application. Next, refer to... Figure 9 The step of updating the position information of the current vertex based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex in step S105 above will be described.

[0116] S901. Based on the position information of each vertex, determine multiple neighboring vertices of the current vertex.

[0117] Specifically, based on the neighborhood relationships between vertices in the position information of each vertex, multiple neighboring vertices of the current vertex are determined.

[0118] S902. Determine the same-pole components based on the position information of each neighboring vertex, the position information of the current vertex, and the preset cotangent weight coefficient.

[0119] Among them, the isotropic component does not distinguish between local geometric structure and feature direction. It is used to apply an equal smoothing correction to all spatial directions of the vertex, thereby uniformly smoothing out the messy high-frequency scanning noise, small bumps and local jitter on the model surface, so that the planar area and rough area can obtain a regular and flat mesh surface effect.

[0120] Specifically, the coordinate difference between the neighboring vertices and the current vertex is calculated, and a weighted average is performed based on the coordinate difference and the preset cotangent weight coefficient. The homogeneous components are then obtained by normalization.

[0121] Specifically, the calculation method for the same-pair components can be shown in the following formula (10): (10) in, Let i be the same-sex component of the i-th vertex in the t-th round. Let i be the set of neighboring vertices of the i-th vertex. The cotangent weighting coefficient is... This represents the position information of the j-th neighboring vertex during the t-th iteration. This represents the position information of the i-th vertex during the t-th iteration.

[0122] S903. Determine the heterogeneous components based on the position information of each neighboring vertex, the position information of the current vertex, and the guiding normal of the current vertex.

[0123] Among them, the anisotropic component is a feature-protective smoothing correction amount, which is used to avoid structural deformation in the normal direction by distinguishing the spatial difference between the surface normal direction and the tangent direction, and only performs slight noise reduction correction in the tangent direction to avoid smoothing and blunting of fine features such as edges, engravings and shallow grooves.

[0124] Specifically, the heterogeneous components can be determined according to the following formula (11): (11) in, Let i be the heterogeneous component of the i-th vertex in the t-th round. Let be the guiding normal of the i-th vertex.

[0125] S904. Update the position information of the current vertex according to the preset filter step size control factor, control operator, same-pair components and opposite-pair components.

[0126] Specifically, the position information of the current vertex can be updated according to the following formula (12): (12) in, This provides the position information of the i-th vertex in the (t+1)-th round. This is the filter step size control factor, with a value ranging from 0 to 1.

[0127] In this embodiment, isotropic and heterotropic components are calculated, and then the position information of the current vertex is updated according to a preset filtering step size control factor, control operator, isotropic and heterotropic components. This enhances the uniformity and smoothness in flat areas, eliminates high-frequency scanning noise in the 3D mesh, and constrains the displacement of the normal direction in complex feature regions, preserving multi-scale fine features such as edges, imprints, and shallow grooves. Simultaneously, the iteration amplitude is controlled based on the filtering step size control factor to avoid mesh topology distortion caused by over-filtering.

[0128] The following describes the step S101 above, which involves determining the radii of multiple virtual spheres based on the position information of each vertex: The global average side length of the model to be denoised is determined based on the position information of each vertex. The global average side length is then amplified using multiple magnification factors to obtain the radii of multiple virtual spheres.

[0129] Specifically, the global average side length can be determined based on the following formula (13): (13) in, The global average side length, This represents the number of topological edges. Let be the Euclidean space length of the m-th topological edge in the model. This Euclidean space length can be determined based on the position information of each vertex.

[0130] Optionally, the number of magnification factors is the same as the number of virtual sphere radii. If the number of virtual sphere radii is 2, then 2 magnification factors are set, and the process of determining the virtual sphere radius can be as follows (14): (14) in, This is the magnification factor.

[0131] In this embodiment, multiple virtual sphere radii are obtained by amplifying the global average side length based on multiple amplification factors, thereby solving the problems of incomplete single-scale feature recognition and poor adaptability of fixed scale.

[0132] Based on the same inventive concept, this application also provides a model denoising device corresponding to the model denoising method. Since the principle of the device in this application is similar to the model denoising method described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0133] Reference Figure 10 The diagram shown is a structural schematic of a model denoising device provided in an embodiment of this application. The device includes: an acquisition module, a first determination module, a second determination module, a third determination module, and an update module; wherein: The acquisition module 1001 is used to acquire the position information of each vertex and the face information of each triangle in the model to be denoised in the current round, and to determine the radius of multiple virtual spheres based on the position information of each vertex. The first determining module 1002 is used to traverse each vertex and, for the current vertex that has been traversed, determine the volume ratio of the current vertex under each virtual sphere radius based on the radius of each virtual sphere, the face information of the multiple triangular facets, and the position information of the current vertex. Each virtual sphere radius indicates a virtual sphere range, and the volume ratio is used to characterize the proportion of the overlap between the virtual sphere range and the model to be denoised in the virtual sphere range. The second determining module 1003 is used to determine the control operator of the current vertex based on the volume ratio of the current vertex under each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius; The third determining module 1004 is used to determine the guiding normal of the current vertex based on the position information of the current vertex, the face information of multiple triangular faces, and the control operator of the current vertex. The update module 1005 is used to update the position information of the current vertex based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex.

[0134] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0135] This application also provides an electronic device, such as... Figure 11 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application, including a processor 1101, a memory 1102, and a bus. The memory 1102 stores machine-readable instructions executable by the processor 1101. When the computer device is running, the processor 1101 and the memory 1102 communicate via the bus, and the processor 1101 executes the machine-readable instructions to perform the processing of the above-mentioned model denoising method.

[0136] This application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the above-described model denoising method.

[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0138] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0139] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A model denoising method, characterized in that, The method includes: Obtain the position information of each vertex and the face information of each triangle in the model to be denoised in the current round, and determine the radii of multiple virtual spheres based on the position information of each vertex; Traverse each vertex. For the current vertex, determine the volume percentage of the current vertex under each virtual sphere radius based on the radius of each virtual sphere, the face information of multiple triangles, and the position information of the current vertex. Each virtual sphere radius indicates a virtual sphere range. The volume percentage is used to characterize the proportion of the overlap between the virtual sphere range and the model to be denoised within the virtual sphere range. The control operator for the current vertex is determined based on the volume percentage of the current vertex under each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius. Based on the position information of the current vertex, the face information of multiple triangles, and the control operator of the current vertex, the guiding normal of the current vertex is determined; The position information of the current vertex is updated based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex.

2. The model denoising method according to claim 1, characterized in that, The step of determining the volume percentage of the current vertex corresponding to each virtual sphere radius based on the radius of each virtual sphere, the facet information of the multiple triangular facets, and the position information of the current vertex includes: Based on the radius of the virtual sphere, determine the initial sampling point of the current vertex within the range of the virtual sphere corresponding to the radius of the virtual sphere; Based on the face information of the multiple triangular facets and the position information of the current vertex, multiple internal sampling points are selected from multiple initial sampling points. The internal sampling points are located inside the model to be denoised. Based on the number of internal sampling points and the number of initial sampling points, the volume percentage of the current vertex corresponding to the radius of the virtual sphere is determined.

3. The model denoising method according to claim 2, characterized in that, The step of determining the initial sampling point of the current vertex within the virtual sphere range corresponding to the virtual sphere radius based on the virtual sphere radius includes: Using the current vertex as the center, construct a bounding box for the current vertex based on the radius of the virtual sphere; A preset number of discrete sampling points are generated within the bounding box of the current vertex; Discrete sampling points whose distance from the current vertex is less than or equal to the radius of the virtual sphere are used as initial sampling points corresponding to the radius of the virtual sphere of the current vertex.

4. The model denoising method according to claim 2, characterized in that, The step of selecting multiple internal sampling points from multiple initial sampling points based on the facet information of the multiple triangular facets and the position information of the current vertex includes: Based on the information of each triangular facet, determine the position of the center point of each triangular facet; Using the current vertex as the topological starting point, a breadth-first search is performed on the mesh surface of the model to be denoised using a pre-constructed topology table to obtain at least one initial triangular facet. Each initial triangular facet is then added to the local manifold facet set of the current vertex, wherein the distance between the center point of each initial triangular facet and the current vertex is less than or equal to the radius of the virtual sphere. Based on the set of local manifold patches and the patch information of each triangular patch in the model to be denoised, multiple internal sampling points are selected from multiple initial sampling points.

5. The model denoising method according to claim 4, characterized in that, The step of selecting multiple internal sampling points from multiple initial sampling points based on the set of local manifold patches and the patch information of each triangular patch in the model to be denoised includes: In the set of local manifold patches, determine the target triangular patch that is closest to each initial sampling point; Based on the center point position of the target triangle, the normal of the triangle in the triangle's information, and the position information of each initial sampling point, determine the discrimination factor of each initial sampling point of the current vertex; The initial sampling point whose discrimination factor is less than or equal to the first preset value is taken as the internal sampling point of the current vertex.

6. The model denoising method according to claim 1, characterized in that, The step of determining the control operator for the current vertex based on the volume percentage of the current vertex under each virtual sphere radius and the weight coefficient corresponding to each virtual sphere radius includes: The absolute value of the difference between the volume percentage corresponding to the radius of each virtual sphere of the current vertex and the second preset value is taken as the feature volume percentage corresponding to the radius of each virtual sphere of the current vertex. The multi-scale features of the current vertex are determined based on the proportion of each feature volume corresponding to each virtual sphere radius of the current vertex and the preset weight coefficients corresponding to each virtual sphere radius. The control operator for the current vertex is determined based on the multi-scale features of the current vertex and the preset sensitivity parameters.

7. The model denoising method according to claim 1, characterized in that, The step of determining the guiding normal of the current vertex based on the position information of the current vertex, the face information of multiple triangles, and the control operator of the current vertex includes: Based on the position information of the current vertex and the face information of multiple triangular faces, determine multiple neighboring faces of the current vertex; The initial normal of the current vertex is determined based on the patch information of multiple neighboring patches; The feature guidance weights are determined based on the control operator, the patch normals in the patch information of multiple neighboring patches, and the initial normal of the current vertex. The guiding normal of the current vertex is determined based on the normals of multiple neighboring faces, the feature guiding weights, and the preset distance weights.

8. The model denoising method according to claim 1, characterized in that, The step of updating the position information of the current vertex based on the control operator of the current vertex, the guiding normal of the current vertex, and the position information of each vertex includes: Based on the position information of each vertex, determine multiple neighboring vertices of the current vertex; Based on the position information of each neighboring vertex, the position information of the current vertex, and the preset cotangent weight coefficient, the same-pole components are determined; Based on the position information of each neighboring vertex, the position information of the current vertex, and the guiding normal of the current vertex, the heterogeneous components are determined; The position information of the current vertex is updated based on the preset filter step size control factor, the control operator, the same-pair component, and the different-pair component.

9. The model denoising method according to claim 1, characterized in that, The process of determining the radii of multiple virtual spheres based on the position information of each vertex includes: Based on the position information of each vertex, determine the global average side length of the model to be denoised; The global average side length is amplified by multiple amplification factors to obtain multiple virtual sphere radii.

10. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when the electronic device is running, are executed by the processor to perform the steps of the model denoising method as described in any one of claims 1 to 9.

Images