Efficient Shape Matching and Retrieval at Multiple Scales

We propose a novel approach for the recognition, clustering and retrieval of shapes. The heart of our methodology is a shape matching algorithm based on dynamic programming, that operates implicitly at multiple scales, but avoids the heavy computational cost of the explicit curvature scale space representation. Our method provides for clustering, visualization and browsing of a data set, as well as for indexing, achieving up to three orders of magnitude speed-up over sequential scanning. We illustrate the application of our method to real two-dimensional static hand gesture data. We also demonstrate the superiority of our approach over traditional approaches to shape matching and retrieval, such as Fourier descriptors, Geometric and Sequential moments. Our evaluation is based on human relevance judgements following a well established methodology from the information retrieval field.