Industrial CT high-precision adaptive surface extraction method based on region growth
By adopting a moving triangle algorithm based on region growth in industrial CT reverse engineering, combining isosurface curvature and second-order gradient information, a high-precision adaptive mesh model is generated, which solves the problems of poor grid quality and low accuracy in traditional methods, and achieves higher quality and higher precision mesh generation.
Patent Information
- Application Number
- CN202510282367.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-11
- Publication Date
- 2025-06-24
AI Technical Summary
In traditional industrial CT reverse engineering, the grid quality generated by the rasterization algorithm is poor, and jagged edges and narrow triangles are prone to appear, and the accuracy is limited. It cannot effectively obtain the high-precision surface characteristics of the model and cannot meet the dimensional measurement application of industrial CT.
Adaptive generation of the grid and high-precision surface points are realized through the isosurface projection of the initial seed point and the search for second-order gradient zero points.
The generated mesh model is of higher quality, and the triangle faces are closer to equilateral triangles, which reduces the number of triangles and vertices, improves data storage and rendering efficiency, and increases the dimensional measurement accuracy by about 20-50μm.
Smart Images

Figure CN120198393A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of industrial CT digital image processing, and is a method for obtaining a high-precision and curvature-driven adaptive grid model based on industrial CT data by region growing. Compared with the grid extracted by the traditional rasterization algorithm, the grid obtained by the present invention has better quality and higher precision. Background Art
[0002] In industrial CT reverse engineering, the traditional rasterized grid generation method has certain limitations. For example, the Marching Cubes (MC) algorithm proposed by William E. Lorensen and Harvey E. Cline in "Marching Cubes: A High Resolution 3D Surface Construction Algorithm" in 1987 can generate surface grids, but the quality of the generated grids is poor, and there are easily jagged edges and surfaces with long and narrow triangles. Although the traditional threshold method ISO50 method can distinguish materials and air, its accuracy is limited and it is seriously affected by CT image artifacts, and it cannot effectively obtain the high-precision surface features of the model, which does not meet the dimensional metrology applications of industrial CT.
[0003] To overcome these drawbacks, a variety of improved methods have been proposed in the prior art, such as the adaptive grid generation method based on octree and the technique of obtaining the threshold based on the maximum inter-class deviation. However, when these methods are actually applied, there are still problems such as high computational complexity, low accuracy, and redundant surface grids. Therefore, there is an urgent need for a grid generation method that can improve the surface accuracy while ensuring the grid quality. The present method uses the marching triangle algorithm based on region growing, which can accurately control the quality of each triangle, avoid the appearance of long and narrow triangles, and obtain an adaptive triangle grid according to the curvature. In addition, the second-order gradient information is used to calculate the accurate surface points, so as to obtain a high-precision grid model that meets the accuracy requirements of industrial CT dimensional metrology. Summary of the Invention
[0004] The object of the present invention is to provide an adaptive grid generation method based on region growing and second-order gradient information, which can generate a high-quality and adaptive grid model in industrial CT reverse engineering. By combining the marching triangle algorithm and using the isosurface curvature and second-order gradient information, the method can effectively improve the accuracy and quality of the grid, reduce the computational complexity and memory consumption at the same time, and obtain an adaptive grid model.
[0005] The technical solution of the present invention is as follows:
[0006] A method for extracting a high-precision adaptive surface of industrial CT based on region growing, the steps are as follows:
[0007] Step 1: Obtain the image data obtained by industrial CT scanning and perform preprocessing, including denoising and filtering.
[0008] Step 2: First, obtain the initial seed points by stepping from a point at the edge of the reconstruction region using the gray value threshold. Then, calculate the gray value and gradient value of the CT image data in Step 1. Project the initial seed points onto the isosurface successively according to the gray value and gradient value, so that the initial seed points are projected along the gradient direction to the position where the second-order gradient zero point is located. Then, expand the initial seed points into regular hexagons along the tangent plane, and use the six newly generated vertices as new seed points to perform the above isosurface projection steps in the same way to obtain six new grid vertices. Generate six initial triangular patches and an initial growth boundary from the initial seed points and the six grid vertices.
[0009] Further, the specific steps of the isosurface projection are as follows:
[0010] Sample the vertex along the normal direction, calculate the second-order gradient value according to the sampled points, and find the zero point of the second-order gradient, that is, the linear interpolation point of the sampling points with different signs. This point is the high-precision surface point, and project the vertex to this point. The second-order gradient calculation formula is improved from the FDK algorithm, and is specifically as follows:
[0011]
[0012] Among them, x, y, and z represent the three-dimensional coordinates of the reconstruction space, i, j, and k represent the derivative orders of the three variables x, y, and z respectively, f represents the CT voxel value; α represents the proportionality factor in the FDK algorithm; S represents the pixel value of the sinogram after being processed by the SL filter; θ represents the rotation angle of the object being scanned; u and v represent the pixel coordinates in the image.
[0013] Step 3: Select a point with the smallest exterior angle on the initial growth boundary, calculate the gradient and the Hessian matrix H of this point, and then calculate the mean curvature:
[0014]
[0015] Among them, is the gradient vector of this point, H(F) is the Hessian matrix, || is the modulus symbol, and Trace(H) is to calculate the trace of the matrix.
[0016] Perform boundary growth according to the mean curvature information, and then use the advancing triangle algorithm for grid growth. The advancing triangle algorithm uses curvature as the weight factor for the growth length to ensure that denser grids are generated in regions with larger curvature, while sparser grids are generated in regions with smaller curvature, so as to realize the adaptive generation of grids.
[0017] Step 4: Mesh Enclosure and Output. Finally, when the growth of triangles along the isosurface ends, the last triangle on the growth boundary is enclosed, and the final mesh model is output.
[0018] Advantages of the present invention: Through experimental verification, for the dimensional metrology application scenario of industrial CT, the mesh model obtained by the present invention can obtain a higher-quality adaptive triangular mesh. The triangular patches are closer to equilateral triangles, with fewer triangles and vertices, which plays an important role in subsequent data storage, rendering, and dimensional fitting. At the same time, the mesh model has higher-precision mesh vertices, and the dimensional accuracy of the model is improved by about 20 - 50 μm compared with the model obtained by the traditional algorithm MC. Description of the Drawings
[0019] Figure 1 It is a schematic diagram of a CT image projection.
[0020] Figure 2 It is a schematic diagram of the expansion of seed points.
[0021] Figure 3 It is a schematic diagram of the quadratic interpolation error.
[0022] Figure 4 It is a two-dimensional schematic diagram of the boundary growth of the marching triangle algorithm.
[0023] Figure 5 It is a schematic diagram of the mesh vertices by gradient projection.
[0024] Figure 6 It is a schematic diagram of the finally generated mesh model.
[0025] Figure 7 It is a detailed view of the generated mesh model. Detailed Embodiments
[0026] To describe the purpose, technical solutions, and advantages of the present invention more clearly, taking the stepped cylinder dataset obtained by cone-beam CT scanning as an example, and combining the accompanying drawings in the embodiments of the present invention, the technical solutions in the present invention will be described completely.
[0027] A high-precision adaptive surface extraction method for industrial CT based on the region growth algorithm, the steps are as follows:
[0028] Step 1: Obtain the image data obtained by industrial CT scanning, as Figure 1 shown, and perform preprocessing, including denoising and filtering, to improve the quality of the data.
[0029] Step 2: First, use the gray value threshold to step from a point at the edge of the reconstruction region to obtain the initial seed points ( Figure 2the central orange point), then calculate the grayscale value and gradient value of the CT image data in step 1, and project the initial seed points onto the isosurface successively according to the grayscale value and gradient value. As Figure 2 shown, project the initial seed points along the gradient direction to the position where the second-order gradient zero point is located. Using the projection map data to calculate the gradient can reduce one linear interpolation compared with the slice map. As Figure 3 shown, make the calculated vertices closer to the real surface. Then expand the initial seed points into regular hexagons along the tangent plane, and use the six newly generated vertices ( Figure 2 the six blue points in) as new seed points to perform the above-mentioned isosurface projection steps in the same way, and obtain six new mesh vertices ( Figure 2 the remaining six orange points in). Generate six initial triangular patches from the initial seed points and the six mesh vertices ( Figure 2 the triangular patches formed by the initial seed point and two adjacent mesh vertices in) and an initial growth boundary ( Figure 2 the hexagon formed by the six mesh vertices in).
[0030] Furthermore, use second-order gradient projection to ensure the accuracy of the mesh vertices. The specific steps of the isosurface projection are as follows:
[0031] Sample the vertices along the normal direction, calculate the second-order gradient value according to the sampled points, and find the zero point of the second-order gradient, that is, the linear interpolation point of the sampling points with different signs. This point is a high-precision surface point, and project the vertices to this point. The second-order gradient calculation formula is as follows:
[0032]
[0033] where x, y, and z represent the three-dimensional coordinates of the reconstruction space, i, j, and k respectively represent the derivative orders of the three variables x, y, and z, f represents the CT voxel value; α represents the scaling factor in the FDK algorithm; S represents the pixel value of the sinogram after processing with the SL filter; θ represents the rotation angle of the object being scanned; u, v represent the pixel coordinates in the image.
[0034] Step 3: Select a point with the smallest exterior angle on the initial growth boundary, calculate the gradient F and Hessian matrix H of this point, and then calculate the mean curvature:
[0035]
[0036] where, is the gradient vector of this point, H(F) is the Hessian matrix, || is the modulus symbol, and Trace(H) is to calculate the trace of the matrix.
[0037] Boundary growth is performed based on the average curvature information; then, the advancing front triangle algorithm is used for mesh generation. The advancing front triangle algorithm uses curvature as the weight factor for the growth length to ensure that denser meshes are generated in areas with larger curvature and sparser meshes are generated in areas with smaller curvature, thus realizing the adaptive generation of meshes.
[0038] Step 4: Mesh closing and output. Finally, when the growth of triangles along the isosurface ends, the last triangle on the growth boundary is closed, and the final mesh model is output.
[0039] The details of the mesh model obtained by this method are as Figure 7 . A size comparison experiment is carried out between the present invention and the MC method. The experimental sample is the Figure 7 aluminum five-step cylinder shown. The comparison index is the deviation between the size of the mesh model obtained by the surface extraction method and the size measured by the coordinate measuring machine. D1 - D5 respectively represent the inner cylindrical surface of the stepped cylinder and the above four outer cylindrical surfaces. The results are as follows in the table:
[0040] Table 1 Comparison results of size deviations
[0041]
[0042] Judging from the results, the present invention has smaller deviations in most sizes compared with the traditional method, that is, it has higher accuracy.
Claims
1. A high-precision adaptive surface extraction method for industrial CT based on region growing, characterized in that: Here are the steps: Step 1: Obtain image data obtained from industrial CT scanning and perform preprocessing, including denoising and filtering; Step 2: First, the grayscale value threshold is used to step from a point at the edge of the reconstruction area to obtain the initial seed point, and then the grayscale value and gradient value of the CT image data in step 1 are calculated, and the initial seed point is projected to the isosurface according to the grayscale value and gradient value in turn, so that the initial seed point is projected to the position of the second-order gradient zero point along the gradient direction, and then the initial seed point is expanded into a regular hexagon along the tangent plane, and the newly generated six vertices are used as new seed points to perform the isosurface projection step in the same way as above, to obtain six new mesh vertices, and the initial six triangular facets and the initial growth boundary are generated by the initial seed point and the six mesh vertices; Step 3: Select a point with the smallest external angle on the initial growth boundary and calculate the gradient of this point And the Hessian matrix H, and then calculate the mean curvature: in, is the gradient vector of the point, H(F) is the Hessian matrix, || is the modulus symbol, and Trace(H) is the trace of the calculated matrix; Boundary growth is performed based on the average curvature information, and then the marching triangle algorithm is used for mesh growth. The marching triangle algorithm uses curvature as a weight factor for the growth length to ensure that a denser mesh is generated in areas with larger curvatures, and a sparser mesh is generated in areas with smaller curvatures, thereby achieving adaptive mesh generation. Step 4: Mesh closure and output. Finally, the triangle grows along the isosurface, the last triangle on the growth boundary is closed, and the final mesh model is output.
2. The high-precision adaptive surface extraction method for industrial CT based on region growing according to claim 1 is characterized in that: In step 2, the specific steps of isosurface projection are as follows: The vertices are sampled along the normal direction, the second-order gradient value is calculated according to the sampling points, and the zero point of the second-order gradient is found, that is, the linear interpolation point of the sampling points with different signs. This point is a high-precision surface point, and the vertex is projected to this point; the second-order gradient calculation formula is improved from the FDK algorithm, as follows: Among them, x, y, z represent the three-dimensional coordinates of the reconstructed space, i, j, k represent the derivative orders of the three variables x, y, z respectively, f represents the CT voxel value; α represents the scale factor in the FDK algorithm; S represents the pixel value of the sinusoidal graph after processing with the SL filter; θ represents the rotation angle of the scanned object; u and v represent the pixel coordinates in the image.
Citation Information
Cited By
Dental 3D printing sub-pixel edge refinement and adaptive slicing method
CN122510462A
