The aim of this paper is to propose a new method of clustering based on spectral clustering and a new kind of similarity. To do this, the original similarity matrix of data is defined so that the number of considerable components in its graph representation be equal to the cluster numbers. Then the main points of clusters and soft points are determined. A one-sided similarity function is utilized to assign soft points and the clusters.Experimental results on benchmark datasets demonstrate the effectiveness of the proposed clustering method.
