A new spectral clustering method based on data histogram
Yunhui Liu, Siwei Luo · 2008
Spectral clustering has become one of the most popular modern clustering algorithms because it is powerful to find structure in data and simple to implement. Commonly used spectral clustering algorithms define the affinity matrix using the widely used Euclidean metric which is simple but may not perform very well in many cases. In this paper, we give a new spectral clustering method revising the similarity matrix by using density information of data set. Such density information is got from data histogram which we call histogram factor. Given two data points, the revised distance measure is the Euclidean distance between the points multiplied by the histogram factor. Experimental results show that the new method can improve the clustering effect much compared to the commonly used methods.