A novel cluster-based image retrieval

A region-based image retrieval is an approach for more efficient and semantically meaningful retrieval using description of image regions. In this paper we present a novel cluster oriented image retrieval system. A cluster is a collection of regions that are coherent in their features. A privileged component of the system is the constructing of cluster signature (CS) and their feature ranking. The CS features are texture, color, and shape of the clusters centroids. Another distinguishing aspect of the system is its integration of a robust unsupervised learning for the detection of region that allows cluster-based search. The system is evaluated using two different databases and proved to provide successful retrieval results supported by precision estimation.