形状匹配方法研究与展望

Research and Perspective on Shape Matching

形状匹配及分类是计算机视觉中的重要问题.近年来,以形状上下文为代表的基于轮廓的形状匹配方法和以奇点图为代表的基于骨架的形状匹配方法获得了长足的发展.本文介绍了形状匹配问题的基本概念,分析了形状匹配问题的难点,按照基于轮廓和基于骨架的分类方法对近年来最新出现的形状表示与形状匹配的方法进行了详尽的介绍,并介绍了基于度量学习的形状检索方法,本文还详细介绍了近年来形状匹配研究领域常用的一些测试数据库,之后对局部形状匹配和形状分类等有潜力的研究方向进行了展望.最后对形状匹配的整体框架及其应用前景进行了总结.

Shape matching and classification are important issues in computer vision. In recent years, contour-based shape matching approaches （e.g., shape context） and skeleton-based shape matching methods （e.g., shock graph） both have a lot of developments. In this paper, we introduce the basic concept of shape matching, give the di？culties of this topic, and provide a detailed review on the most recent approaches about shape representation and matching for both contour-based and skeleton-based methods. We also give a brief introduction about metric learning based shape retrieval. Moreover, we introduce some widely used benchmarks for shape matching in details, along with some hot topics including partial shape matching, shape classification, etc. Finally, this paper concludes with the whole framework of shape matching and the application perspective of this topic.