Class Representative Computation using Graph Embedding and Clustering

Abstract

One of the methods for object recognition is based on graph embedding. By representing objects expressed as graphs into the vector space, this technique makes it possible to use point matching algorithms as opposed to costly graph matching approaches. In this paper, representatives of object classes in the vector space is obtained through graph embedding. To classify a query, instead of using exhaustive search, a more effective way of comparing it to class representatives is employed. Experimental results demonstrate that the proposed work compares favorably to alternative approaches in a set of object recognition experiments.