Content-based object retrieval using maximum curvature points in contour images

In this paper we present a technique for shape similarity estimation and apply it to content-based indexing and retrieval over large image databases. This technique is based on wavelet decomposition and uses polygonal approximation. It uses simple features (aspect ratio, angles, distances from the centroid, distance ratios and relative positions) extracted at high curvature points. The experimental results and comparisons show the performance of the proposed technique. This technique is also suitable to be extended to the retrieval of 3-D objects.