基于多边形拟合的细胞图像特征提取算法  

Shape Descriptors Based on Polygonal Approximation and Its Computer Simulation

作　　者：卢智勇[1] 梁光明[2] 

机构地区：[1]广东机电职业技术学院,广东广州510515 [2]国防科学技术大学,湖南长沙410073

出　　处：《计算机仿真》2009年第11期268-271,共4页Computer Simulation

摘　　要：细胞图像自动检验与识别是现在生物医学工程领域的一个热点问题。在细胞图像识别中传统应用目标的形状特征提取算法存在不足。为了提高图像提取精度,在分析边界直线拟合基础上提出了一种改进的细胞图像特征提取算法,建立基于平滑后边界拟合多边形的曲率向量、边长向量,分析定义了基于拟合多边形的拐点数、端点数、曲率均值、曲率方差、长轴、短轴、长短轴比、对称率、平行度等形状特征描述子。改进算法应用于基本形状与显微镜细胞图像中目标形状的特征提取,计算机仿真表明算法得到的形状描述子可以有效区分基本形状,并具有位移不变,并保证细胞图像提取的精度。The automatic detection and recognition of cell image is a hotspot in biomedical engineering domain. The shape characters of objects are widely applied in the image identification. Based on the analysis of linear fitting for boundary, an improved image extraction algrithm is proposed, and the curvature vector and length vector for smoothed boundary approximate polygon are established. Then shape descriptors such as number of corners, number of peaks, variance of curvature, maximum chord, vertical chord, chord - ratio, symmetry degree, and parallel degree are defined. The algorithm is applied to process basic shape images and cell images. Computer simulation results show that the extracted features can distinguish basic shape effectively and are invariant in translation.

关 键 词：多边形拟合 边界平滑 曲率向量 形状描述子 平行度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]