Integrating contour and skeleton for shape classification

Shape analysis has been a long standing problem in the literature. In this paper, we address the shape classification problem and make the following contributions: (1) We combine both contour and skeleton (also local and global) information for shape analysis, and we derive an effective classifier. (2) We collect a challenging shape database in which there are 20 categories of animals, with each having 100 shapes. All these shapes are obtained from real images with a large variation in pose, viewing angle, articulation, and self-occlusion. (3) We emphasize the importance of having good representation for shape classification to address the unique characteristics of shape. A thorough experimental study is conducted showing significant improvement by the proposed algorithm over many of the state-of-the-art shape matching and classification algorithms, on both our dataset and the well-known MPEG-7 dataset. In addition, we applied our algorithm for recognizing and classifying objects from natural images and obtained very encouraging results.