Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6939894 | Pattern Recognition | 2016 | 12 Pages |
Abstract
Representing shapes in terms of meaningful parts is a fundamental problem in shape analysis and part-based object representation. Decomposition methods typically utilize handcrafted geometric rules in a nondata-driven manner. However, these rules are insufficient to mimic human decomposition behavior, which limits the applications of decomposition in vision tasks. In this paper, we propose a novel shape analysis framework that integrates shape decomposition with shape classification in fundamental level. We first train probabilistic models for contours and part cuts involved in decomposition process. Next, we construct a data structure called “decomposition graph” whose nodes represent intermediate contours and whose edges represent part cut selections. The decomposition and classification results are obtained by efficiently searching the optimal path on decomposition graph with minimum energy. Experimental results show that such integrated framework improves the decomposition performance under various shape deformations and achieves competitive classification performance.
Related Topics
Physical Sciences and Engineering
Computer Science
Computer Vision and Pattern Recognition
Authors
Chun Wang, Zhongyuan Lai,