Shape and texture clustering: Best estimate for the clusters number

The most difficult problem in automatic clustering is the determination of the total number of final clusters NclusterNcluster. In the present paper, a new method for finding NclusterNcluster is proposed and is compared with previously developed methods. The proposed method is based on the minimization of the functional θ(Ncluster)=αNcluster+β∑iNcluster1ni+1Ncluster∑i=1Nclusterdist(Ci) where ni is the number of shapes and textures in cluster CiCi, dist(Ci)dist(Ci) is the intra-cluster distance and αα and ββ are two parameters controlling the grain of the clustering. The proposed method provides almost perfect clustering for the Kimia-25, Kimia-99, MPEG-7 shape databases, subset of Brodatz, full Brodatz and UIUCTex texture databases and provides better results than all previously proposed methods for automatic clustering.