An Invariant Feature Representation for shape Retrieval

Developing suitable feature representation and similarity algorithms is still a challenging problem in shape retrieval system. In this paper, we have proposed an invariant feature representation for shape-based image indexing and retrieval. Using an approximate polygons algorithm, first the given number of dominant points is generated, then transforms (normalizes) each shape to a canonical reference frame, all dominant points are mapped into a linear number of invariant feature points. After all feature points are renumbered beginning with the origin, fixed correspondences are used. Without complex structural matching or high pairwise matching, this relatively simple scheme is effective and robustly tolerates translation, scaling, rotation and inversion ,and increase the speed of shape retrieval without sacrificing accuracy.