Which Representation to Choose for Image Cluster

Abstract

To choose best representation for image cluster, we analyzed the distinguish ability of three typical image representations, color histogram, Gabor texture and geometric moment by fitting image distance curve, and show different retrieval results of image retrieval for different image representations. To depict the distinguish ability of different image representations, we present a cluster discriminant index called Simplified Overall Cluster Quality, which constitutes cluster compaction and cluster separation, and the experiment results show that the image representation with best distinguish ability also possesses maximum discriminant index value. This index can be used to determine optimal representation for clustering or the order of feature used to cluster.

DOI: http://dx.doi.org/10.11591/telkomnika.v12i8.5458