Ki Hyun Nam, Young Shik Moon

The KIPS Transactions:PartB , Vol. 8, No. 2, pp. 215-222, Apr. 2001
10.3745/KIPSTB.2001.8.2.215,   PDF Download:

Abstract

In this paper a new shape feature, referred to as the shape correlogram, is proposed and an efficient method of content-based image retrieval using the shape correlogram is presented. The existing color correlogram improves the efficiency of content based image search by combining the spatial correlation and the color information, but it does not produce a good performance for the images having the similar object shapes with different colors. To resolve this problem the correlogram of edges, referred to as the shape correlogram, is proposed. The shape correlogram represents the correlation of edge angles, while the color correlogram expresses only the correlation in respect to the distance of colors. The shape correlogram is a two-dimensional feature vector with gradient axis and angle axis. Each axis is divided into 24 bins of angles, resulting in 576 elements in total. To reduce the data size of shape correlogram and to maintain the rotation invariance, the projected shape correlogram is also proposed. Experimental results have shown that the proposed retrieval methods based on the shape correlogram and the projected shape correlogram achieve good performance, compared with the existing methods.

Statistics
Cite this article