A hybrid shape descriptor for object recognition

Object recognition via shape matching has been a fundamental topic in robot vision. More and more technologies are widely used in the field of robot and automation in recent years. The shape contour contains meaningful information for object characterization, therefore, an effective representation of shape contour is important for the capability of a shape matching method. In this work, we propose a hybrid shape descriptor which contains salient shape features in different aspects to derive a "rich" descriptor. The hybrid descriptor employs three invariants including: area invariant, arc length invariant and central distance invariant of a given zone in a shape. The three invariants make the description representative and discriminative. Shape matching is explored by calculating shape similarity with the dynamic programming algorithm which is powerful to find the best correspondence between shape contours. The proposed method is invariant to rotation, scale variation, and occlusion. Our method is robust to noise as well. The invariance and robustness of the proposed method are all validated in the extensive experiments. The comparable results on the benchmark datasets indicate that the recognition rate is essentially improved by our method.