Optimum Design of Chamfer Distance Transforms

The distance transform has found many applications in image analysis. The Euclidean distance transform is computationally very intensive, and efficient discrete algorithms based on chamfer metrics are used to obtain its approximations. The chamfer metrics are selected to minimize an approximation error. In this paper, a new approach is developed to find optimal chamfer local distances. This new approach gives smaller approximation error than found in previous work, is easier to visualize, and can be easily extended to chamfer metrics that use large neighborhoods. I. Introduction The distance transform has been applied in many image analysis tasks. It is used for shape analysis, feature detection, skeletonization, segmentation and multiscale morphological filtering [1], [2], [3], [4], [5], [6]. Since the computational cost of the exact Euclidean distance transform is relatively high, several fast algorithms have been developed to approximate it [4], [5]. A major class of such algorithm..