WO2022057250A1 - 三角网格的分割去噪方法 - Google Patents

三角网格的分割去噪方法 Download PDF

Info

Publication number
WO2022057250A1
WO2022057250A1 PCT/CN2021/087447 CN2021087447W WO2022057250A1 WO 2022057250 A1 WO2022057250 A1 WO 2022057250A1 CN 2021087447 W CN2021087447 W CN 2021087447W WO 2022057250 A1 WO2022057250 A1 WO 2022057250A1
Authority
WO
WIPO (PCT)
Prior art keywords
triangular mesh
triangle
denoising
triangular
mesh data
Prior art date
Application number
PCT/CN2021/087447
Other languages
English (en)
French (fr)
Inventor
潘威
戴超凡
鲁学权
Original Assignee
广东奥普特科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 广东奥普特科技股份有限公司 filed Critical 广东奥普特科技股份有限公司
Priority to US18/026,588 priority Critical patent/US20230334627A1/en
Publication of WO2022057250A1 publication Critical patent/WO2022057250A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/70Denoising; Smoothing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/20Finite element generation, e.g. wire-frame surface description, tesselation
    • G06T17/205Re-meshing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/11Region-based segmentation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/23Clustering techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/20Finite element generation, e.g. wire-frame surface description, tesselation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/13Edge detection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/136Segmentation; Edge detection involving thresholding
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/187Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20024Filtering details
    • G06T2207/20028Bilateral filtering

Definitions

  • the present application relates to the technical field of image processing, for example, to a triangular mesh segmentation and denoising method.
  • the triangular mesh model will inevitably be polluted by noise, and the noise will reduce the data quality of the triangular mesh model itself. Moreover, the noise will also affect the processing effect of subsequent meshes, thereby affecting the The effect of 3D scanned images.
  • the denoising algorithms and technologies in the related art have at least the following problems: 1) Most of the triangular mesh denoising methods in the related art use local neighborhood information, and when denoising, the isotropic points or surfaces in the neighborhood are assigned Larger weights, anisotropic points or surfaces are assigned smaller weights, and even smaller weights will still affect the denoising effect, and will destroy sharp features to a certain extent; 2) In order to suppress anisotropic Point or face the influence of the denoising effect, parameters such as angle and distance will be introduced during denoising, which will undoubtedly increase the difficulty of parameter tuning, and this method is still not easy to achieve the expected results, especially those People who are not familiar enough with the algorithm, this tends to increase their burden; 3) Some other methods use more information for mesh denoising, which often involves a lot of computation and usually makes the speed slower.
  • the present application provides a triangular mesh segmentation and denoising method, which can effectively solve the segmentation problem of a noisy model, which not only improves the running speed of noise processing, but also effectively preserves the boundary features and details of the data.
  • An embodiment provides a method for dividing and denoising a triangular mesh, including: reading triangular mesh data including N triangular patches, judging the noise level of the triangular mesh data, and determining the noise level of the triangular mesh data is high. Perform optimization processing on the data of the preset value, where N>1; use the region growing segmentation algorithm to divide the triangular mesh data, so that the triangular mesh data forms multiple sub-regions; use the hole filling algorithm to optimize the divided data. and filtering the segmented triangular mesh data using a denoising algorithm.
  • FIG. 2 is a schematic structural diagram of a side-sharing triangle provided by an embodiment of the present application.
  • FIG. 3 is an effect diagram of triangular mesh data forming multiple sub-regions provided by an embodiment of the present application
  • FIG. 4 is a diagram of the segmentation effect of a small noise according to an embodiment of the present application.
  • FIG. 6 is an effect diagram of marking an edge greater than a D thr value with a small noise under a preset D thr value according to an embodiment of the present application
  • FIG. 7 provides an effect diagram of segmenting a large noise and marking it to the original noise model according to an embodiment of the application
  • FIG. 8 is one of the denoising effect diagrams for a large noise according to an embodiment of the present application.
  • FIG. 9 is the second diagram of the denoising effect on a large noise according to an embodiment of the present application.
  • FIG. 10 is one of the denoising effect diagrams for a small noise according to an embodiment of the present application.
  • FIG. 11 is the second diagram of the denoising effect on a small noise according to an embodiment of the present application.
  • FIG. 13 is a graph of denoising results of a small noise in the case of a hole algorithm and a hole-free algorithm, respectively, according to an embodiment of the present application.
  • the terms “installation”, “connection”, “connection”, “fixation” and other terms should be understood in a broad sense, for example, it may be a fixed connection or a detachable connection, Or integrally connected; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal communication between the two components.
  • installation e.g., it may be a fixed connection or a detachable connection, Or integrally connected; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal communication between the two components.
  • a method for segmenting and denoising a triangular mesh comprising the following steps:
  • the probability density function of Gaussian noise is calculated by It can be seen that the Gaussian noise when ⁇ 0.3 is defined as large noise, where z is the coordinate value of the triangular grid, ⁇ is the local coordinate mean of the triangular grid, and ⁇ is the standard deviation.
  • the data with large noise is optimized by constructing an objective optimization function, so as to enhance the boundary information of the data, and it is more convenient to process the segmentation of the data as a whole.
  • a region growing segmentation algorithm is used to segment the triangular mesh data, so that the triangular mesh data forms multiple sub-regions.
  • the common region growing algorithm usually uses the positional relationship for diffusion, and can use the normal or connection information as the boundary condition for judging the growth.
  • the preset value D thr is set as the judgment condition of the division strength coefficient and the growth boundary.
  • the region growing segmentation algorithm of the segmentation process of the present application concentrates the features and signals of the current region, effectively reducing the influence of signals in different regions, and, in the optimization process, the noisy data is first segmented and then filtered. , and perform denoising processing on the sub-regions after the region segmentation. Compared with the related technology, the boundary features and details are effectively preserved. At the same time, the technical solution of the present application significantly improves the stability of the algorithm. good optimization results can be obtained for the segmentation problem.
  • the threshold D thr is set as the boundary judgment condition for region growth, and the norm
  • F1 is equivalent to a seed surface that has not been divided. It traverses each common triangle of the seed surface F1, and judges whether the value of
  • the normal direction n i of the triangle that needs to be corrected is counted
  • the normal direction n j of the triangle in each area of the two-ring neighborhood S(i) is counted
  • n i and n j are accumulated
  • the calculation method for accumulating with n j is: Wherein, A is the sum of the accumulated number of the cosine value of n i and the cosine value of n j
  • cos(n i , n j ) represents the cosine value of n i and the cosine value of n j .
  • segmentation process of the present application can be embedded in all local point or surface denoising algorithms, which has strong versatility and excellent running performance.
  • the fast normal filtering algorithm, the bilateral normal filtering algorithm, the guided normal filtering algorithm and the L1 median filtering algorithm are respectively used to filter the divided triangular mesh data to obtain the desired image.
  • the filtering process of the fast normal filtering algorithm includes: firstly weighted average of adjacent surface normals, then use threshold to eliminate interfering surfaces, and iteratively filter the noisy surface normals, and then iteratively update the vertex positions to make them consistent with The surface normals after denoising are consistent, so as to achieve image filtering.
  • the filtering process of the bilateral normal filtering algorithm includes: firstly using the spatial distance weight and the method weight, combined with the bilateral operator, iteratively update the normal domain, and then iteratively update the vertex position, so as to realize the filtering of the image.
  • the filtering process of the guided normal filtering algorithm includes: first, using joint bilateral filtering to process the normal direction of the triangular facet, and then updating the position of the vertex according to the filtered normal direction, so as to realize the filtering of the image. Moreover, when performing joint bilateral filtering on the normals of the triangular patches, an appropriate guiding normal should be selected.
  • the filtering process of the L1 median filter algorithm consists of: first, preprocessing the noisy input mesh; then, using the L1 median filter to estimate the normals of the denoised surface; filter.
  • p 1 , p 2 , p 3 , and p 4 are the coordinates of the four vertices of the co-sided triangle, respectively.
  • the coordinates of p 1 are (x 1 , y 1 , z 1 )
  • the coordinates of p 2 The coordinates are (x 2 , y 2 , z 2 )
  • the coordinates of p 3 are (x 3 , y 3 , z 3 )
  • the coordinates of p 4 are (x 4 , y 4 , z 4 )
  • p 1 -p 3 means (x 1 , y 1 , z 1 )-(x 3 , y 3 , z 3 )
  • p 1 -p 4 means (x 1 , y 1 , z 1 )-(x 4 , y 4 , z 4 )
  • p 3 -p 1 means (x 3 , y 3 , z 3 )-(x 1
  • D(e) x represents the differential edge operator of the colateral triangle in the x direction in the 3D Cartesian coordinate system
  • D(e) y represents the differential edge operator of the colateral triangle in the y direction in the 3D Cartesian coordinate system
  • D( e) z represents the differential edge operator for a colateral triangle in the z direction in a 3D Cartesian coordinate system.
  • the model obtained by the L1 median filter algorithm, the next row in Figure 8 represents the denoising effect with the segmentation framework of the present application, and the next row in Figure 8 corresponds to the original noise from the left to the right.
  • model the model obtained after segmentation and then using the bilateral normal filtering algorithm, the model obtained after segmentation and then using the fast normal filtering algorithm, the model obtained after segmentation and then using the guided normal filtering algorithm, and the model obtained after segmentation and then using the L1 median filtering algorithm
  • the obtained model obviously, the image optimization effect obtained by this application is better.
  • the denoising results of the hole filling algorithm and the hole filling algorithm are compared: among them, (a) is the clustering result without the hole filling algorithm. ; (b) denoising based on the clustering results without the hole filling algorithm in (a); (c) based on the results of (b) and adding the hole filling algorithm to optimize the clustering; (d) based on the denoising of (c) Noise results, as can be seen from the figure, the denoising effect of the hole filling algorithm is better.
  • the segmentation algorithm of the present application serves the denoising algorithm, and the purpose is to propose a framework for enhancing the triangular mesh denoising algorithm in the related art, improve the performance of the denoising algorithm, and embed some denoising algorithms into the segmentation of the present application.
  • the influence of non-anisotropic neighborhoods on the denoising results is avoided, the performance of the operation is improved, and the protection of sharp features is effectively and significantly enhanced.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • Computer Graphics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Image Processing (AREA)

Abstract

一种三角网格的分割去噪方法,包括读取包含N个三角面片的三角网格数据,判断所述三角网格数据的噪声等级,并对噪声等级高于预设值的数据进行优化处理;采用区域生长分割算法分割所述三角网格数据,使所述三角网格数据形成多个子区域;采用孔洞填充算法优化分割后的所述三角网格数据;及采用去噪算法对分割后的所述三角网格数据进行滤波。

Description

三角网格的分割去噪方法
本申请要求申请日为2020年9月15日、申请号为202010965873.1的中国专利申请的优先权,该申请的全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像处理的技术领域,例如涉及一种三角网格的分割去噪方法。
背景技术
随着三维扫描数据在人工智能设备制造、工业检测、目标识别、和VR/AR等领域的广泛应用,提高扫描数据的质量对上述领域的发展具有重大意义。在相关技术中,三维扫描得到的初始点云数据量非常多,存储量巨大,不方便进行算法处理,为此,通常需要对初始的数据进行表面重建,得到三角网格模型,再获得简单而常用的数据。
然而,在扫描和重建过程中,发现三角网格模型会不可避免地被噪声污染,噪声会降低三角网格模型本身的数据质量,而且,噪声还会影响到后续网格的处理效果,从而影响三维扫描图像的效果。其中,相关技术中的去噪算法和技术至少存在以下问题:1)相关技术中大多数的三角网格去噪方法使用局部邻域信息,去噪时,邻域内各向同性的点或面赋值较大的权值,各向异性的点或面赋值较小的权值,而再小的权值仍然会影响去噪效果,会在一定程度上破坏尖锐特征;2)为抑制各向异性的点或面对去噪效果的影响,在去噪时会引入角度、距离等参数,这无疑会增大参数调优的难度,而且,这样的方式仍不容易达到预期的结果,特别是那些对算法不够熟悉的人,这往往会加重他们的负担;3)其他一些方法利用更多的信息进行网格去噪,往往涉及大量的计算,通常会使速度更为缓慢。
发明内容
本申请提供了一种三角网格的分割去噪方法,能够有效解决带噪模型的分割问题,既提高了对噪声处理的运行速度,也有效地保留了数据的边界特征和细节。
一实施例提供了一种三角网格的分割去噪方法,包括:读取包含N个三角面片的三角网格数据,判断所述三角网格数据的噪声等级,并对所述噪声等级高于预设值的数据进行优化处理,其中,N>1;采用区域生长分割算法分割所述三角网格数据,使所述三角网格数据形成多个子区域;采用孔洞填充算法优化分割后的所述三角网格数据;及采用去噪算法对分割后的所述三角网格数据进行滤波。
附图说明
图1为本申请一实施例提供的流程图;
图2为本申请一实施例提供的一个共边三角形的结构示意图;
图3为本申请一实施例提供的三角网格数据形成多个子区域的效果图;
图4为本申请一实施例对一个小噪声的分割效果图;
图5为本申请一实施例提供的一个小噪声在不同的D thr值下的分割效果图;
图6为本申请一实施例提供的一个小噪声在预设的D thr值下的对大于D thr值的边进行标记的效果图;
图7为本申请一实施例提供对一个大噪声的分割处理并标记到原噪声模型的效果图;
图8为本申请一实施例对一个大噪声的去噪效果图之一;
图9为本申请一实施例对一个大噪声的去噪效果图之二;
图10为本申请一实施例对一个小噪声的去噪效果图之一;
图11为本申请一实施例对一个小噪声的去噪效果图之二;
图12为本申请一实施例提供的一个小噪声在不同的高斯空间滤波核参数下的去噪效果图;
图13为本申请一实施例提供的一个小噪声分别在有孔洞算法和无孔洞算法的情况下的去噪结果图。
具体实施方式
在本申请的描述中,需要理解的是,术语“上”、“下”、“前”、“后”、“左”、“右”、水平”等指示的方位或位置关系为基于附图所示的方位或位置关系,仅是为了便于描述本申请和简化描述,而不是指示或暗示所指的装置或元件必须具有特定的方位、以特定的方位构造和操作,因此不能理解为对本申请的限制。
在申请中,除非另有明确的规定和限定,术语“安装”、“相连”、“连接”、“固定”等术语应做广义理解,例如,可以是固定连接,也可以是可拆卸连接,或一体地连接;可以是机械连接,也可以是电连接;可以是直接相连,也可以通过中间媒介间接相连,可以是两个元件内部的连通。对于本领域的普通技术人员而言,可以根据具体情况理解上述术语在本申请中的具体含义。
以下结合附图1~12和具体实施例对本申请作进一步详细说明,但不作为对本申请的限定。
一种三角网格的分割去噪方法,包括以下步骤:
S1中,读取包含N个三角面片的三角网格数据,N>1,判断三角网格数据的噪声等级,并对噪声等级δ≥0.3le的数据进行优化处理,le为噪声等级的单位,其中,对于小噪声的数据,直接进入S2。
例如:当读取的三角网格数据仅包含高斯噪声时,由高斯噪声的概率密度函数
Figure PCTCN2021087447-appb-000001
可知,定义σ≥0.3时的高斯噪声为大噪声,其中,z为三角网格的坐标值,μ为三角网格局域坐标均值,σ为标准差。
并且,在S1中,通过构建目标优化函数来对大噪声的数据进行优化,以增强数据的边界信息,更便于对数据整体的分割处理。而该目标优化函数可以为:
Figure PCTCN2021087447-appb-000002
其中,多个三角面片之间形成有共边三角形,
Figure PCTCN2021087447-appb-000003
表示第i个经过优化处理的顶点,p i表示第i个未被优化处理的顶点,
Figure PCTCN2021087447-appb-000004
表示顶点
Figure PCTCN2021087447-appb-000005
和顶点p i的第二范数值,α和β均表示权重系数,w(e)为基于法向的边的权值分布函数,
Figure PCTCN2021087447-appb-000006
表示每个共边三角形的差分边缘运算符D(e)的第二范数值,
Figure PCTCN2021087447-appb-000007
表示三角面片的约束项系数R(e)的第二范数值,R(e)为三角面片的约束项系数,如图2所示,当共边三角形的四个顶点分别为p 1、p 2、p 3、p 4时,R(e)=(p1-p2+p3-p4) 2,从而有效地优化了三角面片的形状,并防止了数据中出现尖刺和翻转三角面片。
S2中,采用区域生长分割算法分割三角网格数据,使三角网格数据形成多个子区域。
其中,普通的区域生长算法通常利用位置关系进行扩散,可以利用法向或连接信息作为判断生长边界条件。但是,对于含噪声的模型,法向的判断非常容易受噪声影响,因此,本方案通过设定预设值D thr作为分割强度系数和生长边界的判断条件。当两个三角形共面的时候,共边三角形的差分边缘运算符D(e)的L2范数(模长)为0,L2范数是指向量每个元素的平方和的平方根。因此,通过边缘运算符的模长,可以提取边界特征。
本申请的分割过程的区域生长分割算法集中了当前区域的特征和信号,有效地降低了不同区域的信号所带来的影响,并且,在优化的过程中,对带噪数 据先分割再进行滤波,并对区域分割后的子区域分别进行去噪处理,相比于相关技术,有效地保留了边界特征和细节,同时,本申请的技术方案显著地提高了算法的稳定性,在处理噪声模型的分割问题时能够获得很好的优化效果。
S2-1中,提取三角网格数据中的一个三角面片,遍历三角面片的每个共边三角形,计算每个共边三角形的差分边缘运算符D(e)的值;
S2-2中,将阈值D thr设定为区域生长的边界判断条件,对每个共边三角形计算范数||D(e)||,使满足||D(e)||<D thr关系式的共边三角形与三角面片划分为同一个子区域。
如图3所示,F1相当于一个未被分割处理的种子面,遍历种子面F1的每个共边三角形,并判断||D(e)||值是否小于D thr,若满足||D(e)||<D thr,则标记为新种子面F2,再不断地进行扩散和聚类,直到所有的新种子面的||D(e)||值都大于D thr
并且,由于噪声的影响,S2中,将三角网格模型分割成多个区域后,可能会存在一些局部分割错误“小区域”,此时的局部“小区域”网格将严重影响去噪效果,而利用区域孔洞填充算法的精细化处理并进行优化可以得到更精准的分割结果。
S3中,采用孔洞填充算法优化分割后的三角网格数据;
S3-1中,对多个子区域进行检索,获得需要进行更正的三角形;
S3-2中,遍历每个需要进行更正的三角形的二环邻域S(i);
S3-3中,统计需要进行更正的三角形的法向n i,统计二环邻域S(i)的每个区域的三角形的法向n j,对n i和n j进行累加,对n i和n j进行累加的计算方法为:
Figure PCTCN2021087447-appb-000008
其中,A为n i的余弦值与n j的余弦值进行累加的数量和,cos(n i,n j)表示n i的余弦值与n j的余弦值。
此外,本申请的上述分割过程可以嵌入到所有局域点或面的去噪算法,其 通用性强,运行性能优异。
S4中,分别采用快速法向滤波算法、双边法向滤波算法、引导法向滤波算法和L1中值滤波算法对分割后的三角网格数据进行滤波,从而获得所需的图像。
其中,快速法向滤波算法的滤波过程包括:先通过对相邻面法线进行加权平均,然后利用阈值排除干扰面,并迭代过滤带噪声的面法线,再迭代更新顶点位置,使之与去噪后的面法线一致,从而实现图像的滤波。
双边法向滤波算法的滤波过程包括:先利用空间距离权重和方法权重,结合双边算子,迭代更新法向域,然后再迭代更新顶点位置,从而实现图像的滤波。
引导法向滤波算法的滤波过程包括:首先,采用联合双边滤波处理三角面片的法向,然后中根据滤波后的法向更新顶点的位置,从而实现图像的滤波。并且,在对三角面片的法向进行联合双边滤波处理的时候,应选定了一个合适的引导法向。
L1中值滤波算法的滤波过程包括:首先,对有噪声的输入网格进行预处理;然后,使用L1中值滤波器估算去噪表面的法线;最后,迭代更新顶点位置,从而实现图像的滤波。
而且,共边三角形的差分边缘运算符D(e)的计算方式为:
Figure PCTCN2021087447-appb-000009
其中,p 1、p 2、p 3、p 4分别为共边三角形的四个顶点坐标,在三维直角坐标系中,p 1的坐标为(x 1,y 1,z 1),p 2的坐标为(x 2,y 2,z 2),p 3的坐标为(x 3, y 3,z 3),p 4的坐标为(x 4,y 4,z 4),p 1-p 3表示(x 1,y 1,z 1)-(x 3,y 3,z 3),p 1-p 4表示(x 1,y 1,z 1)-(x 4,y 4,z 4),p 3-p 1表示(x 3,y 3,z 3)-(x 1,y 1,z 1),p 2-p 1表示(x 2,y 2,z 2)-(x 1,y 1,z 1),p 3-p 2表示(x 3,y 3,z 3)-(x 2,y 2,z 2),p 4-p 3表示(x 4,y 4,z 4)-(x 3,y 3,z 3),Δ 1,2,3为p 1、p 2和p 3所围成的三角形面积,Δ 1,3,4为p 1、p 3和p 4所围成的三角形面积。
并且,范数||D(e)||的计算方式为:
Figure PCTCN2021087447-appb-000010
其中,D(e) x表示三维直角坐标系中x方向的共边三角形的差分边缘运算符,D(e) y表示三维直角坐标系中y方向的共边三角形的差分边缘运算符,D(e) z表示三维直角坐标系中z方向的共边三角形的差分边缘运算符。
如图4~5所示,一个噪声等级δ=0.2le的模型直接采用S2进行分割,且分割效果随分割强度系数D thr变化而变化。并且,由图6可知,通过比较||D(e)||和D thr的大小,可以寻找和提取出该模型的边界,以更利于后续的去噪。
如图7所示,以一个噪声等级δ=0.4le的大噪声模型为例,先对该模型进行优化处理,再进行分割,然后通过区域孔洞填充的精细化处理,最后再标记到原噪声模型。
如图8所示,以一个噪声等级δ=0.8le的cube形状模型为例,该模型的噪声过大,则先进行预处理,增强边界信息,再进行分割,将分割聚类情况标记到原始噪声模型,再进行去噪,其中,UNF表示快速法向滤波算法,BNF表示双边法向滤波算法,及GNF表示引导法向滤波算法,并且,图8中的上一行表示相关技术的原始方法的去噪效果,图8中的上一行从左到右依次为原始模型、采用双边法向滤波算法获得的模型、采用快速法向滤波算法获得的模型、采用引导法向滤波算法获得的模型和采用L1中值滤波算法获得的模型,图8中的下一行表示的是本申请的带有分割框架的去噪效果,图8中的下一行从左到右依 次为将分割结果对应到原始噪声的模型、经分割再采用双边法向滤波算法获得的模型、经分割再采用快速法向滤波算法获得的模型、经分割再采用引导法向滤波算法获得的模型和经分割再采用L1中值滤波算法获得的模型,显然,本申请所获得的图像优化效果更好。
如图9所示,以一个噪声等级δ=0.5le的vase模型为例,通过上一行和下一行的对比可知,带有分割框架的去噪效果更好,并且,图9中的英文字母的含义与图8相同。
如图10所示,以一个噪声等级δ=0.1le的fandish模型为例,通过上一行和下一行的对比可知,带有分割框架的去噪效果更好,并且,图10中的英文字母的含义与图8相同。
如图11所示,以一个噪声等级δ=0.2le的dodecahedron模型为例,通过上一行和下一行的对比可知,带有分割框架的去噪效果更好,并且,图11中的英文字母的含义与图8相同。
如图12所示,以一个噪声等级δ=0.2le的dodecahedron模型为例,通过上一行和下一行的对比可知,当采用引导法向滤波算法时,在不同的高斯空间滤波核λr中,对于不同的参数λr,带有分割框架的去噪效果受参数的影响更小,优化的效果更好。
如图13所示,以一个噪声等级δ=0.1le的octahedron模型为例,对比有孔洞填充算法和无孔洞填充算法的去噪结果:其中,(a)为没有加入孔洞填充算法的聚类结果;(b)基于(a)中没有加入孔洞填充算法的聚类结果进行去噪;(c)基于(b)的结果加入了孔洞填充算法进行优化聚类;(d)基于(c)的去噪结果,从图可知,有孔洞填充算法的去噪效果更佳。
本申请的分割算法是为去噪算法服务,目的是提出一种增强相关技术中的 三角网格去噪算法的框架,提高去噪算法性能,并且,将一些去噪算法嵌入到本申请的分割模型中,避免了非异性邻域对去噪结果的影响,提高了运行的性能,从而有效地显著增强了对尖锐特征的保护。

Claims (8)

  1. 一种三角网格的分割去噪方法,包括:
    读取包含N个三角面片的三角网格数据,判断所述三角网格数据的噪声等级,并对所述噪声等级高于预设值的数据进行优化处理,其中,N>1;
    采用区域生长分割算法分割所述三角网格数据,使所述三角网格数据形成多个子区域;
    采用孔洞填充算法优化分割后的所述三角网格数据;及
    采用去噪算法对分割后的所述三角网格数据进行滤波。
  2. 如权利要求1所述的三角网格的分割去噪方法,其中,所述采用区域生长分割算法分割所述三角网格数据,使所述三角网格数据形成多个子区域包括:
    提取所述三角网格数据中的一个三角面片,遍历所述三角面片的每个共边三角形,计算每个共边三角形的差分边缘运算符D(e)的值;及
    将阈值D thr设定为区域生长的边界判断条件,对每个共边三角形计算范数||D(e)||,使满足||D(e)||<D thr关系式的共边三角形与所述三角面片划分为同一个子区域。
  3. 如权利要求2所述的三角网格的分割去噪方法,其中,所述共边三角形的差分边缘运算符D(e)的计算方式为:
    Figure PCTCN2021087447-appb-100001
    其中,p 1、p 2、p 3、p 4分别为所述共边三角形的四个顶点坐标,在三维直角坐标系中,p 1的坐标为(x 1,y 1,z 1),p 2的坐标为(x 2,y 2,z 2),p 3的坐标为(x 3,y 3,z 3),p 4的坐标为(x 4,y 4,z 4),p 1-p 3表示(x 1,y 1,z 1)-(x 3,y 3,z 3),p 1-p 4 表示(x 1,y 1,z 1)-(x 4,y 4,z 4),p 3-p 1表示(x 3,y 3,z 3)-(x 1,y 1,z 1),p 2-p 1表示(x 2,y 2,z 2)-(x 1,y 1,z 1),p 3-p 2表示(x 3,y 3,z 3)-(x 2,y 2,z 2),p 4-p 3表示(x 4,y 4,z 4)-(x 3,y 3,z 3),Δ 1,2,3为p 1、p 2和p 3所围成的三角形面积,Δ 1,3,4为p 1、p 3和p 4所围成的三角形面积。
  4. 如权利要求2所述的三角网格的分割去噪方法,其中,所述范数||D(e)||的计算方式为:
    Figure PCTCN2021087447-appb-100002
    其中,D(e) x表示三维直角坐标系中x方向的共边三角形的差分边缘运算符,D(e) y表示三维直角坐标系中y方向的共边三角形的差分边缘运算符,D(e) z表示三维直角坐标系中z方向的共边三角形的差分边缘运算符。
  5. 如权利要求1所述的三角网格的分割去噪方法,其中:所述采用孔洞填充算法优化分割后的所述三角网格数据包括:
    对多个子区域进行检索,获得需要进行更正的三角形;
    遍历每个需要进行更正的三角形的二环邻域S(i);及
    统计需要进行更正的三角形的法向n i,统计所述二环邻域S(i)的每个区域的三角形的法向n j,对所述n i和所述n j进行累加。
  6. 如权利要求5所述的三角网格的分割去噪方法,其中,所述对所述n i和所述n j进行累加的计算方法为:
    Figure PCTCN2021087447-appb-100003
    其中,所述A为所述n i的余弦值与所述n j的余弦值进行累加的数量和,所述cos(n i,n j)表示所述n i的余弦值与所述n j的余弦值。
  7. 如权利要求1所述的三角网格的分割去噪方法,其中,所述对噪声等级高于噪声等级δ≥0.3le的数据进行优化处理,其中,le为噪声等级的单位。
  8. 如权利要求1所述的三角网格的分割去噪方法,其中:所述采用去噪算法对分割后的所述三角网格数据进行滤波包括分别采用快速法向滤波算法、双边 法向滤波、引导法向滤波和L1中值滤波算法对分割后的所述三角网格数据进行滤波。
PCT/CN2021/087447 2020-09-15 2021-04-15 三角网格的分割去噪方法 WO2022057250A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/026,588 US20230334627A1 (en) 2020-09-15 2021-04-15 Method for segmenting and denoising triangle mesh

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010965873.1 2020-09-15
CN202010965873.1A CN112085750A (zh) 2020-09-15 2020-09-15 一种三角网格的分割去噪方法

Publications (1)

Publication Number Publication Date
WO2022057250A1 true WO2022057250A1 (zh) 2022-03-24

Family

ID=73737941

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/087447 WO2022057250A1 (zh) 2020-09-15 2021-04-15 三角网格的分割去噪方法

Country Status (3)

Country Link
US (1) US20230334627A1 (zh)
CN (1) CN112085750A (zh)
WO (1) WO2022057250A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115409950A (zh) * 2022-10-09 2022-11-29 卡本(深圳)医疗器械有限公司 一种面绘制三角网格的优化方法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112085750A (zh) * 2020-09-15 2020-12-15 广东奥普特科技股份有限公司 一种三角网格的分割去噪方法
CN112529811A (zh) * 2020-12-17 2021-03-19 中国地质大学(武汉) 一种保留地形表面结构特征的dem数据去噪方法
CN115294258B (zh) * 2022-09-26 2022-12-23 腾讯科技(深圳)有限公司 三维模型的展开方法、装置、设备及计算机可读存储介质
CN116630330A (zh) * 2023-07-26 2023-08-22 征图新视(江苏)科技股份有限公司 一种基于边缘差分的三角网格平面缺陷检测方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105741355A (zh) * 2016-02-01 2016-07-06 华侨大学 一种三角网格模型的块分割方法
CN109345627A (zh) * 2018-09-26 2019-02-15 华侨大学 一种三角网格模型特征保持混合简化方法
US20190258225A1 (en) * 2017-11-17 2019-08-22 Kodak Alaris Inc. Automated 360-degree dense point object inspection
CN110930334A (zh) * 2019-11-26 2020-03-27 浙江大学 一种基于神经网络的网格去噪方法
CN112085750A (zh) * 2020-09-15 2020-12-15 广东奥普特科技股份有限公司 一种三角网格的分割去噪方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107045732B (zh) * 2016-02-06 2020-05-19 高德软件有限公司 数字地面模型化简方法及装置
CN108898558A (zh) * 2018-06-13 2018-11-27 天津大学 一种保特征网格的曲面去噪方法
CN110497727B (zh) * 2019-08-28 2020-12-01 华侨大学 一种立体石雕加工的最优加工空间选取方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105741355A (zh) * 2016-02-01 2016-07-06 华侨大学 一种三角网格模型的块分割方法
US20190258225A1 (en) * 2017-11-17 2019-08-22 Kodak Alaris Inc. Automated 360-degree dense point object inspection
CN109345627A (zh) * 2018-09-26 2019-02-15 华侨大学 一种三角网格模型特征保持混合简化方法
CN110930334A (zh) * 2019-11-26 2020-03-27 浙江大学 一种基于神经网络的网格去噪方法
CN112085750A (zh) * 2020-09-15 2020-12-15 广东奥普特科技股份有限公司 一种三角网格的分割去噪方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115409950A (zh) * 2022-10-09 2022-11-29 卡本(深圳)医疗器械有限公司 一种面绘制三角网格的优化方法
CN115409950B (zh) * 2022-10-09 2024-02-06 卡本(深圳)医疗器械有限公司 一种面绘制三角网格的优化方法

Also Published As

Publication number Publication date
CN112085750A (zh) 2020-12-15
US20230334627A1 (en) 2023-10-19

Similar Documents

Publication Publication Date Title
WO2022057250A1 (zh) 三角网格的分割去噪方法
Haralick Edge and region analysis for digital image data
CN109685080B (zh) 基于霍夫变换与区域生长的多尺度平面提取方法
Bithell A classification of disease mapping methods
US8224089B2 (en) Information processing device and information processing method, image processing device and image processing method, and computer program
CN105806765A (zh) 一种显微ct扫描土体空间孔隙结构的精细化表征方法
CN112164145B (zh) 基于点云数据快速提取室内三维线段结构的方法
CN109472267A (zh) 图片相似度比对算法
Branch et al. Automatic hole-filling of triangular meshes using local radial basis function
CN106815842A (zh) 一种改进的基于超像素的图像显著性检测方法
CN112785591B (zh) 一种ct影像中肋骨骨折的检测与分割方法及装置
WO2022227486A1 (zh) 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质
CN108280833B (zh) 一种植物根系分叉特征的骨架提取方法
CN107256571A (zh) 一种基于深度学习与自适应差分盒的分形维数估计方法
CN117635609B (zh) 一种塑胶产品生产质量视觉检测方法
Chang et al. Automatic classification of lidar data into ground and non-ground points
CN114862715A (zh) 一种融合地形特征语义信息的tin渐进加密去噪方法
Branch et al. A hole-filling algorithm for triangular meshes using local radial basis function
CN113899349B (zh) 海浪参数检测方法、设备及存储介质
CN113409332B (zh) 一种基于三维点云的建筑物平面分割方法
Sapirstein Segmentation, reconstruction, and visualization of ancient inscriptions in 2.5 D
CN115937151B (zh) 一种作物叶片卷曲程度的判定方法
CN112541897A (zh) 一种基于人工智能的面瘫程度评估系统
CN113780421B (zh) 基于人工智能的脑部pet影像识别方法
CN112529811A (zh) 一种保留地形表面结构特征的dem数据去噪方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21868088

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21868088

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 01.08.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 21868088

Country of ref document: EP

Kind code of ref document: A1