Semi-Supervised Clustering of Corner-Oriented Attributed Graphs

Abstract

This paper describes a new algorithm for image semi-supervised clustering. In particular, the proposed approach introduces corner-oriented attributed graphs(COAG) constructed based on modified Harris corner extraction method to represent structure objects . 2D-Laplacianface is used to reduce the dimension of feature matrix obtained from COAG. Feature vector is built just from the output of dimensionality reduction. This vector denotes the input to the classifier. Semi-supervised k-mean clustering method (S2KMCM) is carried out as semi-clustering method. Experimental results show that COAG can preserve the structure information of image and S2KFCM can be applied to both clustering and classification tasks by labeled and unlabeled data together.