Shape similarity retrieval under affine transforms

The maxima of curvature scale space (CSS) image have already been used to r epresent 2-D shapes in different applications. The representation has shown robustness under the similarity transformations. Scaling, orientation chan ges, translation and even noise can be easily handled by the representation and its associated matching algorithm. In this paper, we examine the robus tness of the representation under general affine transforms. We have a data base of 1100 images of marine creatures. The contours in this database demo nstrate a great range of shape variation. A database of 5000 contours has b een constructed using 500 real object boundaries and 4500 contours which ar e the affine transformed versions of real objects. The CSS representation i s then used to find similar shapes from this prototype database. The result s provide substantial evidence of stability of the CSS image and its contou r maxima under affine transformation. The method is also evaluated objectiv ely through a large classified database and its performance is compared wit h the performance of two well-known methods, namely Fourier descriptors and moment invariants. The CSS shape descriptor has been selected for MPEG-7 s tandardization. (C) 2001 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.