Point cloud boundary segmentation method and system for tooth color three-dimensional digital model

Abstract

The invention relates to a point cloud boundary segmentation method and system for a tooth color three-dimensional digital model, and the method comprises the steps: carrying out the gamma correction of a tooth HSV color model, and obtaining a point cloud with the local contrast enhancement of a V channel of an original point cloud HSV color model, and a seed point as a boundary line; obtaining and calculating the similarity between each field point and the seed point according to the feature information of field points in a certain field range of the seed point, and taking the points of which the similarity is greater than a preset similarity threshold as boundary points of the boundary line; after performing quality evaluation on the boundary points of the boundary line, dynamically adjusting the weight of gamma correction until the boundary points of the boundary line under the optimal quality evaluation are obtained; and obtaining a closed boundary line according to the finally obtained boundary line points. The method fully reduces noise interference by fusing multiple features such as color, normal and curvature, effectively improves the accuracy and stability of boundary segmentation in combination with dynamic weight, reduces the cost of manual parameter adjustment, and is suitable for point cloud data under different illumination and geometric conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of oral medicine, and in particular to a point cloud boundary segmentation method and system for a colored three-dimensional digital model of teeth. Background Art

[0002] Traditionally, a plaster model of teeth created in the mouth typically requires a professional to draw boundary lines on the model using a red or black marker (a colored marker). These boundary lines are usually closed, irregular curves. A 3D scanning device then scans the plaster model to create a color 3D digital model of the teeth. If algorithms can identify these manually drawn boundary lines on the model, this provides accurate and crucial data for the next step of digital tooth arrangement.

[0003] In the existing technology, the color space (such as HSV, LAB) threshold segmentation method is directly used to extract the marked line area, but the existing segmentation method has some problems: ① Due to uneven illumination during three-dimensional imaging, color information is easily distorted: traditional gamma correction uses fixed parameters and cannot adapt to local brightness changes, resulting in overexposure or underexposure; ② The complex geometric structure of teeth makes segmentation difficult: single features (such as color or curvature) are easily affected by noise, resulting in missegmentation; ③ The fading or color diffusion of the manually marked lines is affected. The current manual parameter adjustment is too subjective and inefficient, making it difficult to accurately distinguish the marked color area from the similar interfering background area. Summary of the Invention

[0004] In order to solve the technical problems mentioned in the background technology, the present invention provides a point cloud boundary segmentation method and system for a colored three-dimensional digital model of a tooth.

[0005] The object of the present invention can be achieved by the following technical solution: A method for segmenting the point cloud boundary of a colored three-dimensional digital model of a tooth, comprising the steps of:

[0006] S100, after performing gamma correction on the tooth HSV color model, obtaining a point cloud with enhanced local contrast of the V channel of the original point cloud HSV color model;

[0007] S200, taking points in the gamma-corrected point cloud whose color features are greater than the set HSV three-component thresholds as seed points of the boundary line;

[0008] S300, obtaining feature information of each point in the point cloud, including at least curvature feature information of the point, normal feature information of the point, and color feature information of the point;

[0009] S400, obtaining and calculating the similarity between each domain point and the seed point based on feature information of domain points within a certain domain range of the seed point, and taking points with similarities greater than a preset similarity threshold as boundary points of the boundary line;

[0010] S500, after performing quality assessment on the boundary points of the obtained boundary line, dynamically adjusting the weight of the gamma correction until the boundary points of the boundary line corresponding to the optimal quality assessment are obtained;

[0011] S600: Obtain a closed boundary line according to the seed point and the boundary points of the boundary line finally obtained.

[0012] Based on the above, in step S500, after obtaining the optimal gamma correction weight, the weight of the similarity calculation in step S400 is dynamically adjusted until the boundary point of the boundary line corresponding to the optimal similarity is obtained, and the boundary point of the boundary line corresponding to the optimal quality assessment and the optimal similarity is used as the boundary point of the boundary line finally obtained.

[0013] Based on the above, in step S300, the curvature calculation of a point includes the following steps:

[0014] S311. For a point, take the points within the k1 radius of the point as the local neighborhood points of the point;

[0015] S312, calculating the covariance matrix of the neighborhood points of the point;

[0016] S313, after performing eigenvalue decomposition on the covariance matrix, three eigenvalues ​​are obtained;

[0017] S314. Further calculate the curvature value of the point based on the three eigenvalues ​​obtained in step S313.

[0018] Based on the above, in step S300, the calculation of the normal feature of the point includes the following steps:

[0019] S321. For a point, the two points closest to the point are taken as the local plane points of the point;

[0020] S322, establishing a plane through the point and the two local plane points;

[0021] S323. Taking the normal of the plane as the normal of the point;

[0022] S324. The cosine value of the angle between the normal of the point and the normal of the seed point is used as the normal angle of the point.

[0023] Based on the above, in step S400, the similarity calculation formula between the neighborhood point of the seed point and the seed point is:

[0024]

[0025] Among them, Similarity represents the similarity between the current point and the seed point; is the color similarity, H j 、S j 、V j are the values ​​of the three components of the HSV color model of the seed point, H j+1 、S j+1 、V j+1 are the values ​​of the three components of the HSV color model of the current point, w H 、w S 、w V They correspond to the weights of H, S, and V in the HSV color model respectively; is the normal similarity, n j is the normal vector of the seed point, n j+1 is the normal vector of the current point; C seed is the curvature value of the seed point; C current is the curvature value of the current point; α and β are weight parameters, Var(d color ) represents the color variance of the points in the k1 neighborhood of the current point; Std(θ) is the standard deviation of the normal angle between the current point and the seed point; w2 and w4 are weight coefficients.

[0026] Based on the above, in step S500, the quality assessment formula is:

[0027] U(B)=w2Var(d color )+w3∑δ(c)+w4Std(θ)

[0028] δ(c): count +1 when |C(p)-C0|>k2

[0029] Among them, U(B) is the quality value of the boundary point of the boundary line; w2, w3, and w4 are weight values ​​respectively; C(p) is the curvature value of point p; δ(c) is the curvature mutation count; C0 is the preset curvature value of the initial boundary line; and k2 is the preset value.

[0030] Based on the above, in step S500, dynamically adjusting the weight of gamma correction includes the following steps:

[0031] S511, randomly selecting a gamma correction weight value w1, and performing gamma correction on the original point cloud;

[0032] S512, after obtaining the boundary points of the point cloud after gamma correction in step S511, obtaining the corresponding quality assessment results;

[0033] S513. After adjusting the weight value w1 upward or downward to a new value, recalculate the quality assessment result U(B). If the value of U(B) increases, adjust the weight value of w1 in the gamma correction in the opposite direction until the value of U(B) reaches the minimum value or is less than the set iteration threshold. Then, end the iteration. At this time, the weight value of w1 in the gamma correction reaches the optimal setting.

[0034] Based on the above, when dynamically adjusting the weight of similarity calculation, construct the objective function J1:

[0035]

[0036] Among them, w5 is the weight value; Similarity mean Similarity is the mean of the similarity of the boundary points of the segmented boundary line; max is the maximum similarity of the boundary points of the segmented boundary line; Indicates the degree of similarity between the boundary points of the segmented boundary line and the seed points;

[0037] By iteratively adjusting the values ​​of the weights w2, w3, and w4, when the value of the objective function J1 reaches the minimum value or is less than the set iteration threshold, the corresponding weight value at this time is the optimal weight value of w2, w3, and w4, and the corresponding similarity is the optimal similarity.

[0038] Based on the above, in step S600, a closed boundary line is generated by a convex hull algorithm according to the seed point and the boundary points of the boundary line finally obtained.

[0039] A point cloud boundary segmentation system for a colored three-dimensional digital model of teeth includes a point cloud boundary segmentation unit. The point cloud boundary segmentation unit uses any of the above-mentioned point cloud boundary segmentation methods for a colored three-dimensional digital model of teeth to perform point cloud boundary segmentation.

[0040] The beneficial effects of the present invention are as follows:

[0041] 1. This invention effectively improves the accuracy of boundary point segmentation by integrating multiple features such as color, normal, and curvature. The calculation of color variance and standard deviation of normal angle enables the algorithm to reduce noise and inconsistent interference while ensuring segmentation accuracy, effectively avoiding missegmentation. Similarity calculation weighs the influence of different features to ensure that boundary points can be accurately identified and segmented. The combination of multiple features and similarity methods makes the point cloud segmentation process more accurate and stable, and can handle different types of point cloud data.

[0042] 2. By dynamically adjusting the gamma correction parameters, it is possible to maximize the boundary quality while preserving as much detail as possible in the original point cloud data, avoiding information loss caused by over-correction. By iteratively adjusting the quality assessment value, it is possible to adaptively find the optimal gamma correction parameters, avoiding the subjectivity of manual parameter adjustment, and can adapt to point cloud data under different lighting conditions;

[0043] 3. By constructing an objective function and combining the weights of color confusion and similarity, the segmented boundary lines can be fine-tuned to ensure that the segmented points are closer to the real boundaries. The dynamic weight adjustment mechanism gives the present invention a strong adaptability. According to different point cloud data and actual application scenarios, the algorithm can automatically adjust the parameters, thereby always maintaining a good segmentation effect in different data sets or complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below with reference to the accompanying drawings;

[0045] Figure 1 It is a schematic diagram of the collaborative framework of the prediction model of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] like Figure 1 As shown in the figure, a method for segmenting the boundary of a point cloud for a colored 3D digital model of a tooth is described. After obtaining the colored 3D digital model of the tooth, gamma correction is applied to the color channels of the original point cloud in the model. Geometric features are then extracted from the gamma-corrected point cloud model, resulting in a gamma-corrected point cloud and model feature information such as curvature and normals. Furthermore, seed points are selected based on the color information of the gamma-corrected points.

[0048] Specifically:

[0049] First, gamma correction is used for processing: the color format of the collected colored point cloud is RGB. After converting the point cloud color model in RGB format to the HSV color model, gamma correction is applied to the brightness (V channel) channel in the HSV color model. First, the brightness variance of the spherical neighborhood with a fixed radius R1 of each point in the point cloud is calculated, and the gamma value is obtained based on the brightness variance calculation. The gamma transform is applied based on the calculated gamma value, thereby achieving local contrast enhancement of the brightness channel of the point cloud. In this embodiment, the size of the fixed neighborhood is adjusted and set according to the actual situation. The gamma correction calculation formula is:

[0050]

[0051] Where: γ is the calculated gamma value; w1 is the weight parameter; is the variance of the brightness of all points in a fixed neighborhood in the point cloud; V i is the brightness value of the V channel of the i-th point in the neighborhood; μ V is the average brightness value of the V channel in the neighborhood; n is the number of points in the neighborhood. The gamma transform calculation formula is:

[0052]

[0053] Among them: I γ is the brightness value of the point cloud after gamma transformation; I corrected is the brightness value before gamma correction; It is the power operation in the gamma transform formula, which is used to correct brightness.

[0054] By performing a power transformation on the brightness, the local contrast of the image is enhanced, thus achieving the effect of gamma correction. Finally, the local contrast enhancement of the brightness channel is completed, and the point cloud after applying gamma correction to the V channel of the HSV color model of the original point cloud is obtained.

[0055] For the point cloud after gamma correction, by setting the threshold of the three HSV components, seed points are screened out. The seed points are points whose color features are greater than the set threshold. For example, if the points on the boundary line of the current dental mold are red, then based on the color features of the red points, by setting the threshold of the three HSV components, red points whose color features are greater than the threshold are screened out, and the screened out points are used as seed points. At this time, the points corresponding to the red manual boundary line in the point cloud actually have two parts, one part is the red seed points, and the other part is the red points close to the edge of the boundary line, but their red feature values ​​are less than the set threshold. For the convenience of description and distinction, these boundary line points are referred to as boundary line boundary points below.

[0056] The existing algorithm is used to extract geometric features from the gamma-corrected point cloud: for the gamma-corrected point cloud, the eigenvalues ​​of the covariance matrix of the local neighborhood of each point are first calculated to finally obtain the curvature of each point. The curvature value calculation process for a point is as follows: S311, for a point, with k1 as the size, the points within the k1 radius of the point are regarded as the local neighborhood points of the point; S312, the covariance matrix of the local neighborhood points of the point is calculated; S313, the covariance matrix is ​​decomposed into eigenvalues ​​to obtain three eigenvalues; S314, based on the three eigenvalues, the curvature value of the point is further calculated. In this embodiment, the size of the neighborhood k1 value is adjusted and set according to the actual situation.

[0057] The process of calculating the normal angle of a point is as follows: S321. For a point, at least two points closest to the point (if the relationship between the two points and the point is special and a plane cannot be determined, then continue to select the third point relatively closest to the point) are used as the local plane points of the point; S322. Since three points can determine a plane, a plane is established through the point and the two local plane points; S323. After obtaining the normal vector of the plane (using the curvature information based on the point cloud surface normal or the direction method of the bounding box normal in the existing technology, the normals of all points on a point cloud object can be made to face outward), the normal vector of the plane is used as the normal vector of the point; S324. The cosine value of the angle between the normal of the point and the normal of the seed point is used as the normal angle of the point.

[0058] In the gamma-corrected point cloud, the color feature value of the HSV three channels of each point is used as the color feature of the point.

[0059] Finally, the curvature features, normal angle features, and color features of each point in the point cloud after gamma correction are calculated.

[0060] According to the color model of the point cloud after gamma correction, the color weight and geometric feature weight are calculated. Further, based on the corrected color information and geometric feature information, the similarity between each area point and the seed point in a certain area of ​​each seed point is calculated. By setting the similarity threshold, the boundary line points are segmented.

[0061] Specifically, based on the gamma-corrected point cloud, the color variance of the gamma-corrected point cloud is further calculated. The calculation process is as follows: First, the variance of the three HSV channels of each point is calculated based on the neighborhood size. Then, the mean of the variance of the three channels of the point is calculated, and the mean is used as the color variance value of the current point. For each point, the variance of the curvature of the point is calculated based on the neighborhood size and the curvature value of each point. The above calculation methods are all existing methods. Among them, the size of the neighborhood value when calculating the variance can be adjusted and set according to actual conditions.

[0062] The point cloud after gamma correction, the color feature information of each point (the value of the HSV color model, color variance), the normal vector, the curvature value and other information are obtained. Then, based on the selected seed point, the similarity between each area point within a certain range of the seed point in the point cloud and the seed point is calculated. By setting a similarity threshold, the points with similarity calculation results greater than the similarity threshold are used as boundary points of the boundary line. At this time, since the number of other points is large, and only a small part of these other points belong to the boundary points of the boundary line, therefore, in order to reduce the amount of calculation and improve the operation efficiency, in this embodiment, the four domain points of each seed point are obtained, that is, the four neighboring points closest to the seed point (by calculating the Euclidean distance between points, the four points closest to the seed point are taken as the points in the four neighborhoods of the seed point). There are actually three possibilities for these four domain points, namely, seed points, boundary points or non-boundary line points. In order to further reduce the amount of calculation, for a point among these four domain points, if the domain point is a seed point, it will not participate in the similarity calculation; if the similarity between the domain point and the corresponding seed point is greater than the preset threshold after calculation, it will be regarded as a boundary point; if the similarity between the domain point and the corresponding seed point is less than the preset threshold after calculation, it will still be regarded as other points. After the similarity calculation of the 4 field points of a seed point is completed, the 4 field points of the second seed point are calculated. At this time, in order to further reduce the amount of calculation, if a point in the 4 field points of the second seed point is a confirmed boundary point, then the boundary point will no longer participate in the similarity calculation of the second seed point. After traversing and calculating each seed point and its associated 4 field points, all the boundary points finally confirmed and obtained are recorded as the boundary points of the boundary line. In this embodiment, the similarity calculation is performed on the 4 field points of each seed point after calculating the feature information of all points. Although some computing power will be wasted, the efficiency of calculating the similarity will be higher after calculating the feature information in advance. In other embodiments, the feature information of each field point and the corresponding seed point can be calculated when calculating the similarity of each field point, and then the similarity is calculated. This will avoid calculating the feature information of a large number of points, but will reduce the efficiency of the similarity calculation. In reality, the size of the field for calculating the similarity can be set according to the actual situation, such as 6 field points, 8 field points, etc. The detailed calculation process is as follows:

[0063] First, the weight parameter is calculated by the color variance of the points in the neighborhood of each point and the standard deviation of the normal angle of each point:

[0064]

[0065] Where: Var(d color ) represents the color variance of points in the neighborhood, reflecting the fluctuation of color changes; Std(θ) is the standard deviation of the normal angle; w2 and w4 are weight coefficients; α and β are weight parameters respectively.

[0066] When the color variance Var(d color ) increases, indicating color confusion, and the color weight α needs to be reduced to avoid over-reliance on noisy data. When the normal angle standard deviation Std(θ) increases, it indicates inconsistent normal directions, and the normal weight β needs to be reduced to reduce normal interference.

[0067] The similarity between a seed point and its domain points is calculated as follows:

[0068]

[0069] Among them: Similarity indicates the similarity between the seed point and the current point. is the color similarity, which indicates the difference between the seed point and the current point in the color space (HSV space). It is calculated by weighted Euclidean distance and indicates the degree of difference between the two points in the HSV color space. j 、S j 、V j are the values ​​of the three components of the HSV color model of the seed point, H j+1 、S j+1 、V j+1 are the values ​​of the three components of the HSV color model of the current point, w H 、w S 、w V They correspond to the weights of H, S, and V in the HSV color model respectively; is the normal similarity, which is the cosine value of the angle between the normal vectors of the seed point and the current point. The closer it is to 1, the more similar the normals are; n j is the normal vector of the seed point, n j+1 is the normal vector of the current point; C seed is the curvature value of the seed point; C current is the curvature value of the current point.

[0070] In this embodiment, the similarity threshold is adjusted and set according to the actual situation; H 、w S 、w V The size of the weight value is also adjusted and set according to the actual situation.

[0071] Beneficial Effects: By considering multi-dimensional features such as color, normals, and curvature, the accuracy of boundary point segmentation is effectively improved. Calculating the standard deviation of color variance and normal angles enables this method to reduce noise and inconsistencies while maintaining segmentation accuracy. Similarity calculations balance the influence of different features to ensure accurate identification and segmentation of boundary points. These calculations make the point cloud segmentation process more accurate and stable, and capable of handling different types of point cloud data.

[0072] Evaluation: First, the quality of the generated boundary points is evaluated. Then, based on the quality evaluation results and the differences between the point cloud model before and after gamma correction, the parameters set in the gamma transformation are dynamically adjusted to achieve the best gamma correction effect. Based on the quality evaluation results, the degree of color confusion of the segmented boundary points, and the similarity of each point, the weight parameters used in the calculation of the similarity of each point in the point cloud are dynamically adjusted to ultimately achieve the best boundary points.

[0073] Specifically: Based on the segmented boundary points, the color of the boundary points and other information, the quality of the generated boundary points is evaluated. The quality value calculation process is as follows:

[0074] U(B)=w2Var(d color )+w3∑δ(c)+w4Std(θ)

[0075] δ(c): count +1 when |C(p)-C0|>k2

[0076] Where: U(B) is the boundary quality value; w2, w3, and w4 are weight values ​​respectively; C(p) is the curvature value of point p; δ(c) is the curvature mutation count, which counts when the curvature of the point is greater than the set threshold; C0 is the preset curvature of the initial boundary line; the size of the k2 value is adjusted and set according to the actual situation.

[0077] Beneficial effect: By comprehensively considering the color difference Var(d color ), curvature mutation δ(c), and the standard deviation Std(θ) of the angles at the boundary points can be used to quantify the quality of the boundary. This assessment helps ensure that the segmented boundary line accurately reflects the true shape of the object. Detecting curvature mutations can help identify critical areas of the boundary line where significant changes often occur. This can avoid missegmentation or over-smoothing.

[0078] In addition, the boundary quality assessment result U(B) provides a target value for the subsequent gamma correction. By continuously optimizing this target, the final segmentation quality of the gamma-corrected point cloud data can be improved.

[0079] Specifically: After obtaining the boundary quality assessment results, the gamma correction parameters are first dynamically adjusted according to U(B). Combined with the degree of change in the V channel in the HSV color channel of the point cloud color, the quality of the boundary points is optimized while minimizing the change in the original data. When the quality value of U(B) reaches the minimum, the gamma correction effect is optimal and the amount of original data lost is minimized. The original color distribution is protected by difference constraints to prevent subsequent algorithm misjudgments caused by correction. The point cloud data under different lighting conditions is adapted to obtain the preprocessed point cloud. There is the following adjustment process: S511, randomly select a value as the weight value of w1 in gamma correction, apply gamma correction to the original point cloud, and obtain the corrected image; S512, perform the above calculation process, and obtain the U(B) quality evaluation result under the current gamma-corrected w1 weight; S513, based on the initially selected gamma-corrected w1 weight value, select a new gamma-corrected w1 weight value up or down, and then recalculate the size of the U(B) value. If the U(B) value increases, adjust the gamma-corrected w1 weight value in the opposite direction until the U(B) value reaches the minimum value or is less than the set iteration threshold, and then end the iteration. At this point, the weight value of w1 in the gamma correction reaches the optimal setting, the U(B) value reaches the minimum, the boundary point quality is the highest, and the most original information is retained. The calculation formula is:

[0080]

[0081] in: is the value of the weight parameter w1 at the current iteration number t; is the value of the weight parameter w1 at the next iteration t+1; Δ is the step size; sign(·) is the sign function; U (t) is the boundary quality evaluation value at the current iteration t; U candidate To use candidate weights The calculated new quality evaluation value. The iteration threshold is set based on the actual situation.

[0082] The current weight Add a step size Δ to get the candidate weight Compare the current quality with the candidate quality U candidate , determine whether to continue adjusting or reversely adjust the value of w1, and adjust w1 according to the result of the sign function to move the weight toward a better quality direction.

[0083] Beneficial effects: By dynamically adjusting the gamma correction parameter w1, it is possible to maximize the boundary quality while retaining as much detail as possible in the original point cloud data, avoiding information loss due to over-correction. By iteratively adjusting the U(B) quality value assessment, the algorithm can adaptively find the optimal gamma correction parameters, avoiding the subjectivity of manual parameter adjustment, and can adapt to point cloud data under different lighting conditions. Gamma correction can adjust the brightness and contrast of the point cloud, making the manual boundary lines on the colored three-dimensional digital model of the tooth clearer, further improving the segmentation effect. Dynamic adjustment allows the algorithm to automatically optimize under different data sets and conditions, improving overall performance.

[0084] After obtaining the point cloud after gamma correction with the optimal weight parameters, we further construct the objective function J1 and iteratively adjust the values ​​of the weights w2, w3, and w4 to minimize the value of the objective function J1. When the value of J1 reaches the minimum or is less than the set iteration threshold, the points of the segmented boundary line are optimal and closest to the points of the true boundary line. The calculation formula is:

[0085]

[0086] Among them: J1 is the constructed objective function; w5 is the weight value; Similarity mean is the mean of the similarity of the segmented boundary points; Similarity max is the maximum similarity of the segmented boundary points; Indicates the degree of similarity between the segmented boundary points and the seed points. Smaller values ​​indicate more similarity between the boundary points and the seed points; larger values ​​indicate less similarity between the boundary points and the seed points. The iteration threshold can be adjusted based on actual conditions.

[0087] Beneficial Effects: By constructing the objective function J1 and incorporating weights for color confusion and similarity, the segmented boundary can be fine-tuned to ensure that the segmented points are closer to the true boundary. Minimizing the value in J1 effectively improves the accuracy of the segmentation results and reduces the risk of missegmentation. By comprehensively considering multiple factors such as color confusion and point similarity, the algorithm can more accurately identify similar or related points during segmentation. This results in more precise segmentation and reduces the probability of missegmentation. The algorithm's dynamic weight adjustment mechanism provides strong adaptability. It can automatically adjust parameters based on different point cloud data and actual application scenarios, maintaining consistently good segmentation results across different datasets and complex scenes. This makes the algorithm effective in processing a wide range of point cloud data types. By optimizing color confusion and improving similarity judgment, the algorithm can reduce the misclassification of points with similar colors or similar geometric features into different segmentation groups. This optimization is particularly beneficial for processing point cloud data in complex scenes, effectively reducing segmentation errors.

[0088] The quality of point cloud processing has been significantly improved by improving segmentation accuracy, enhancing adaptability, reducing erroneous segmentation, improving computational efficiency, and enhancing the practical application value of point cloud data.

[0089] Finally, based on the seed points and the segmented boundary points, the existing convex hull algorithm is used to generate a closed boundary. The convex hull algorithm generates a minimum convex polygon that contains all the seed points and boundary points. This polygon can be used as an approximation of the closed boundary, thus obtaining a closed boundary line.

[0090] This embodiment also provides a point cloud boundary segmentation system for a colored three-dimensional digital model of teeth, including a point cloud boundary segmentation unit. The point cloud boundary segmentation unit uses any of the above-mentioned point cloud boundary segmentation methods for a colored three-dimensional digital model of teeth to perform point cloud boundary segmentation.

[0091] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A point cloud boundary segmentation method for a colored three-dimensional digital model of a tooth, characterized by: Including steps: S100, after performing gamma correction on the tooth HSV color model, obtaining a point cloud with enhanced local contrast of the V channel of the original point cloud HSV color model; S200, taking points in the gamma-corrected point cloud whose color features are greater than the set HSV three-component thresholds as seed points of the boundary line; S300, obtaining feature information of each point in the point cloud, including at least curvature feature information of the point, normal feature information of the point, and color feature information of the point; S400, obtaining and calculating the similarity between each domain point and the seed point based on feature information of domain points within a certain domain range of the seed point, and taking points with similarities greater than a preset similarity threshold as boundary points of the boundary line; S500, after performing quality assessment on the boundary points of the obtained boundary line, dynamically adjusting the weight of the gamma correction until the boundary points of the boundary line corresponding to the optimal quality assessment are obtained; S600: Obtain a closed boundary line according to the seed point and the boundary points of the boundary line finally obtained.

2. The method for segmenting a point cloud boundary of a colored three-dimensional digital tooth model according to claim 1, characterized in that: In step S500, after obtaining the optimal gamma correction weight, the weight of the similarity calculation in step S400 is dynamically adjusted until the boundary point of the boundary line corresponding to the optimal similarity is obtained, and the boundary point of the boundary line corresponding to the optimal quality assessment and the optimal similarity is used as the boundary point of the boundary line finally obtained.

3. The method for segmenting a point cloud boundary of a colored three-dimensional digital tooth model according to claim 1, characterized in that: In step S300, the curvature calculation of a point includes the following steps: S311. For a point, take the points within the k1 radius of the point as the local neighborhood points of the point; S312, calculating the covariance matrix of the neighborhood points of the point; S313, after performing eigenvalue decomposition on the covariance matrix, three eigenvalues ​​are obtained; S314. Further calculate the curvature value of the point based on the three eigenvalues ​​obtained in step S313.

4. The method for segmenting a point cloud boundary of a colored three-dimensional digital model of a tooth according to claim 1, characterized in that: In step S300, the calculation of the normal feature of the point includes the following steps: S321. For a point, the two points closest to the point are taken as the local plane points of the point; S322, establishing a plane through the point and the two local plane points; S323. Taking the normal of the plane as the normal of the point; S324. The cosine value of the angle between the normal of the point and the normal of the seed point is used as the normal angle of the point.

5. The method for segmenting a point cloud boundary of a colored three-dimensional digital tooth model according to claim 1, characterized in that: In step S400, the similarity calculation formula between the neighborhood point of the seed point and the seed point is: Among them, Similarity represents the similarity between the current point and the seed point; is the color similarity, H j 、S j 、V j are the values ​​of the three components of the HSV color model of the seed point, H j+1 、S j+1 、V j+1 are the values ​​of the three components of the HSV color model of the current point, w H 、w S 、w V They correspond to the weights of H, S, and V in the HSV color model respectively; is the normal similarity, n j is the normal vector of the seed point, n j+1 is the normal vector of the current point; C seed is the curvature value of the seed point; C current is the curvature value of the current point; α and β are weight parameters, Var(d color ) represents the color variance of the points in the k1 neighborhood of the current point; Std(θ) is the standard deviation of the normal angle between the current point and the seed point; w2 and w4 are weight coefficients.

6. The method for segmenting a point cloud boundary of a colored three-dimensional digital tooth model according to claim 1, characterized in that: In step S500, the quality assessment formula is: <h2 style=";text-align:left;direction:ltr">U(B)=w2Var(d<h2 style=";text-align:left;direction:ltr"> color <h2 style=";text-align:left;direction:ltr"> )+w3∑δ(c)+w4Std(θ) δ(c): count +1 when |C(p)-C0|>k2 Among them, U(B) is the quality value of the boundary point of the boundary line; w2, w3, and w4 are weight values ​​respectively; C(p) is the curvature value of point p; δ(c) is the curvature mutation count; C0 is the preset curvature value of the initial boundary line; and k2 is the preset value.

7. The method for segmenting a point cloud boundary of a colored three-dimensional digital tooth model according to claim 1, characterized in that: In step S500, dynamically adjusting the weight of gamma correction includes the following steps: S511, randomly selecting a gamma correction weight value w1, and performing gamma correction on the original point cloud; S512, after obtaining the boundary points of the point cloud after gamma correction in step S511, obtaining the corresponding quality assessment results; S513. After adjusting the weight value w1 upward or downward to a new value, recalculate the quality assessment result U(B). If the value of U(B) increases, adjust the weight value of w1 in the gamma correction in the opposite direction until the value of U(B) reaches the minimum value or is less than the set iteration threshold. Then, end the iteration. At this time, the weight value of w1 in the gamma correction reaches the optimal setting.

8. The method for segmenting a point cloud boundary of a colored three-dimensional digital tooth model according to claim 2, characterized in that: When dynamically adjusting the weight of similarity calculation, construct the objective function J1: Among them, w5 is the weight value; Similarity mean Similarity is the mean of the similarity of the boundary points of the segmented boundary line; max is the maximum similarity of the boundary points of the segmented boundary line; Indicates the degree of similarity between the boundary points of the segmented boundary line and the seed points; By iteratively adjusting the values ​​of the weights w2, w3, and w4, when the value of the objective function J1 reaches the minimum value or is less than the set iteration threshold, the corresponding weight value at this time is the optimal weight value of w2, w3, and w4, and the corresponding similarity is the optimal similarity.

9. The method for segmenting a point cloud boundary of a colored three-dimensional digital tooth model according to claim 1, characterized in that: In step S600 , a closed boundary line is generated using a convex hull algorithm based on the seed points and the boundary points of the boundary line finally obtained.

10. A point cloud boundary segmentation system for a colored three-dimensional digital model of a tooth, comprising a point cloud boundary segmentation unit, characterized in that: The point cloud boundary segmentation unit performs point cloud boundary segmentation using the point cloud boundary segmentation method of the tooth color three-dimensional digital model according to any one of claims 1 to 9.

Images