A novel method of contour extraction based on dynamic programming

Geometric contour, as an important object feature, has been used for shape matching and object recognition. Recently Mortensen and Barrett have proposed a method for contour extraction. Based on Dijkstra's 2-D dynamic programming, the method has a number of advantages over previous approaches. However, the drawbacks of this method are its interactive requirement during the optimal search and high computational complexity. In this paper, we propose a new optimal search strategy to avoid the interactive requirement by allowing the variations of the control points and integrate a multi-resolution framework to reduce computational complexity. Experiments show the effectiveness of our proposed approach.