A NEW SIMILARITY MEASURE USING HAUSDORFF DISTANCE MAP (MP-L3)

Author(s) : Etienne Baudrier (CReSTIC, France) Gilles Millon (CReSTIC, France) Frédéric Nicolier (CReSTIC, France) Su Ruan (CReSTIC, France)

Abstract : Shape dissimilarity measure is a hot topic. A great deal of work has been done, on the one hand, on very simple shapes (convex objects or one-piece object without holes etc...) based on geometric descriptors and on the other hand, on complex shapes (colored and textured ones) based on intensity measures. In this paper, we are presenting a new dissimilarity measure based on the Hausdorff distance (HD). The comparison process is structured as follows: a morphological multiresolution analysis is applied to the two images in order to produce low resolution approximations. Secondly a distance map is constructed at each scale by the computation of the HD restricted through a sliding-window. A signature is then extracted from the distance map and a statistic model is used to describe the signature distribution. Our method eventually allows to take a reliable measure of the image dissimilarity. As an application, the algorithm has been successfully tested on an ancient illustrations database which contains complex binary shapes.

Menu