AN EFFICIENT CONTOUR-BASED LAYERED SHAPE DESCRIPTOR FOR IMAGE RETRIEVAL

This paper introduces a novel contour-based shape descriptor for the retrieval of simple shape as well as complex shape images. In this method, the shape points are partitioned into three layers according to their distance to the center of shape distribution. For each layer, we calculate the distance, angle and pixel ratio as the features of a shape. The possible combinations of the features generate various descriptors, and we evaluate and compare their performance with that of curvature scale space (CSS) and angular radial transform (ART) descriptors, which are adopted by MPEG-7 Visual experimentation Model. The method provides good invariant properties in rotation, scaling and their combination. Experimental results show that the proposed method is better than CSS and ART descriptors in most tests. Furthermore, the new descriptor is much more compact and computational efficiency than CSS and ART for shape image retrieval.