MULTI-SCALE REPRESENTATION AND OPTIMAL MATCHING OF NON-RIGID SHAPES

In this paper, an alternative and more compact form of the Multi-Scale Convexity Concavity (MCC) representation [1] for matching non-rigid shapes with a single contour is proposed, along with an optimization framework designed to improve its efficiency. In the original MCC representation, a feature vector containing a measure of convexity/concavity at multiple scale-levels was constructed for each contour point on an object’s shape. Although such a rich descriptor exhibited improved performance over existing techniques, the inherent redundancy between scale levels implied high storage requirements. In this paper, we discuss possibilities for reducing the redundancy of this representation. In addition, we propose an iterative optimization framework for determining the relative proportions in which information from different contour point features should be combined in the final similarity measure. The experiments conducted using the MPEG-7 shape database show that such optimization can further improve retrieval performance of the proposed MCC method.