Object-based image retrieval using dominant color pairs between adjacent regions

Abstract

Most existing methods for content-based image retrieval handle an image as a whole, instead of focusing on an object of interest. This paper proposes object-based image retrieval based on the dominant color pairs between adjacent regions. From a segmented image, the dominant color pairs between adjacent regions are extracted to produce color adjacency matrix, from which candidate regions of 1313 images are selected. The similarity measure between the query image and candidate regions in DB images is computed based on the color correlogram technique. Experimental results show the performance improvement of the proposed method over existing methods.