基于图像中物体骨架形状特征的目标匹配

Target Matching Based on Object Skeleton Shape Features in the Image

针对在形状匹配过程中相似外形的不同类目标不易区分的问题,本文采用骨架形状对目标进行描述和相似性度量。与以往使用物体的外轮廓作为形状特征进行匹配的方法不同,本文改用图像的骨架形状作为形状特征来进行描述,采用经典的形状上下文作为形状描述的匹配算法。提取图像区域的骨架后,建立与原图像一一对应的图像骨架数据库,进行目标与数据库之间的相似性度量。与直接用外部轮廓进行形状匹配的方法相比,采用骨架形状匹配的方法可以更好的分辨出相似外形的不同类目标,增大类间区分度,可有效降低相似外形目标进行匹配时的误匹配几率。

For the problem that the different types of objects with similar shape are difficult to distinguish in the process of shape matching, this paper uses the combination of skeleton feature and shape context operator to perform target matching.Different from the previous method of using the outer contour of the object as the shape feature, the skeleton shape of the image is used as the shape feature to describe, and the classic shape context is used as the shape description matching algorithm. After extracting the skeleton of the image region, an image skeleton database corresponding to the original image is established, and the similarity measure between the target and the database is performed.Compared with the method of directly matching the outer contour with shape, the method of skeleton shape matching can better distinguish different types of targets with similar shapes, and increase the discrimination between classes, which can effectively reduce the mismatch probability when similar shape targets are matched.