基于骨架树描述符匹配的物体相似性度量方法  

OBJECTS SIMILARITY MEASURE BASED ON SKELETON TREE DESCRIPTOR MATCHING

作　　者：刘文予[1] 刘俊涛[1] 

机构地区：[1]华中科技大学电子与信息工程系图像信息处理与智能控制教育部重点实验室,湖北武汉430074

出　　处：《红外与毫米波学报》2005年第6期432-436,共5页Journal of Infrared and Millimeter Waves

基　　金：国家自然科学基金(60273099)资助项目

摘　　要：提出了通过匹配骨架树来度量物体相似度的新方法.该方法用一种新的物体特征—树描述符来表示一个骨架树,用搜索树描述符中最长公共子串的方法获得最大同构子树;用新的模型度量骨架枝的形状相似性,并把形状特征和拓扑特征的有机的结合起来.最后,对骨架有环和骨架上有噪声的情况进行了讨论.实验结果表明,本方法计算的相似度具有物体的旋转、大小、平移不变性,算法时间复杂度为o(n3),n为物体骨架枝数,对多种物体取得了令人满意的结果.A new object feature, i.e. , tree descriptor, was used to represent a skeleton tree. And the maximal isomorphic subtree was obtained by searching the longest same substring in the tree descriptors, then a novel method of objects similarity measuring based on matching skeleton tree was presented. During the matching process, a new model was used to measure the similarity between two skeleton branches, furthermore, the shape feature and topology feature were associated perfectly. The problems of cyclic skeleton and noise on the skeleton were discussed. The similarity computing with our method is invariant under rotation, scaling, translation, and the time complexity of the algorithm is o（ n^3）, n is the number of the skeleton branches in the skeleton tree, and the experiments on a variety of objects get satisfying results.

关 键 词：骨架树 树描述符 相似性 匹配 计算机视觉 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]