Circle Filling Rate Descriptor for Object Recognition

Abstract

Research on shape recognition has shown that a number of recognition approaches in the literature are sensitive to rotation. In this work, we present a shape description technique, which is based on the local shape filling rate of object silhouettes. After drawing circles of increasing radius around each skeleton point, we compute the ratio between pixels that lie within the shape and the total number of pixels in each circle. Experimental evaluation of the proposed approach including a comparison with the previous technique demonstrates effectiveness of our algorithm for a shape retrieval experiment.