Contour Based Shape Matching for Object Recognition

Haoran Xu, Jianyu Yang, Zhanpeng Shao, Yazhe Tang, Youfu Li

Abstract: To improve computational efficiency and solve the problem of low accuracy caused by geometric transformations and nonlinear deformations in the shape-based object recognition, a novel contour signature is proposed. This signature includes five types of invariants in different scales to obtain representative local and semi-global shape features. Then the Dynamic Programming algorithm is applied to shape matching to find the best correspondence between two shape contours. The experimental results validate that our methods is robust to rotation, scaling, occlusion, intra-class variations and articulated variations. Moreover, the superior shape matching and retrieval accuracy on benchmark datasets verifies the effectiveness of our method.