非封闭内腔CT图像局部多尺度凸包分割方法  

Local multi-scale convex hull segmentation method of non-closed inner cavity CT images

作　　者：秦倩 邹永宁[1,2] 黄业凌 韦会鸿 王俊瑶 Qin Qian;Zou Yongning;Huang Yeling;Wei Huihong;Wang Junyao(College of Optoelectronic Engineering,Chongqing University,Chongqing 400044,China;Engineering Industrial Computed Tomography Nondestructive Testing of the Ministry of Education,Chongqing University,Chongqing 400044,China)

机构地区：[1]重庆大学光电工程学院,重庆400044 [2]重庆大学工业CT无损检测教育部工程研究中心,重庆400044

出　　处：《仪器仪表学报》2024年第11期233-242,共10页Chinese Journal of Scientific Instrument

基　　金：国家重点研发计划(2022YFF0706400)项目资助。

摘　　要：从工业CT图像上分割精密零件内腔区域对于零件的尺寸测量具有重要意义。零部件内腔通常是不封闭的,其CT灰度与背景灰度相近,利用现有图像分割算法无法准确分割出内腔。提出一种结合凸包思想和数学形态学的局部多尺度凸包算法,在初始分割基础上对图像进行内腔填充,再通过闭运算和布尔操作实现完整内腔的分割。经过多种分割方法对比,实验结果表明局部多尺度凸包算法在汽车零部件CT图像上的F1分数达到了0.9735。所提算法正确性较高,能够快速、准确地分割出不同类型工业CT图像中的非封闭内腔区域。Segmenting the inner cavity regions of precision parts from industrial CT images is crucial for accurate dimensional measurements.However,inner cavities are often non-closed and exhibit CT grayscale values similar to the background,making accurate segmentation challenging for existing algorithms.To address this,this paper introduces a local multi-scale convex hull algorithm that integrates convex hull concepts with mathematical morphology.Starting from an initial segmentation,the algorithm fills the inner cavity regions,followed by closing and Boolean operations to achieve complete segmentation.Comparative experiments with various segmentation methods demonstrate the effectiveness of the proposed approach,achieving an F1 score of 0.9735 on CT images of automotive parts.The results indicate that the proposed algorithm offers high accuracy and efficiency,enabling the precise and rapid segmentation of non-closed inner cavity regions in diverse industrial CT image applications.It is of great significance to segment the inner cavity region of precision parts from industrial CT images for dimensional measurement of the inner cavity.The inner cavities of parts are usually not closed,and their CT grayscale is similar to the background grayscale,making it difficult to accurately segment the inner cavity using existing image segmentation algorithms.This paper proposes a local multi-scale convex hull algorithm that combines the idea of convex hulls and mathematical morphology.Based on the initial segmentation,the algorithm fills the inner cavity of the image,and then achieves complete segmentation of the inner cavity through closing operations and Boolean operations.After comparing various segmentation methods,experimental results show that the F1 score of the local multi-scale convex hull algorithm on automotive part CT images reached 0.9735.The algorithm proposed in this paper has a high degree of correctness and can quickly and accurately segment non-closed inner cavity areas in various types of industrial CT images.

关 键 词：CT图像 非封闭内腔 图像分割 局部多尺度凸包算法 

分 类 号：TH741[机械工程—光学工程]