Similarity Matrix And Spectral Clustering

Spectral clustering is a newly proposed method which has been intensely studied in recent years. It bases directly on similarity matrix, and uses Spectral Graph Theory to form clusters. Similarity matrix is key point of spectral clustering. Different similarity matrixes have great influence on the results segmented by spectral clustering. The research of the influence of similarity matrix to spectral clustering and how to construct similarity matrixes best suit spectral clustering is very meaningful. Many studies have analyzed the spectral clustering on both numerical and image data. However, few papers are focusing the similarity matrix and its influence to spectral clustering.This paper concentrated on this problem and gave experimental result about it. First this paper briefly discussed the construction of the similarity matrix, different similarity matrix based on different distance functions and different similarity based on different integration methods. Then it implemented the designed experiment to analysis the influence of these different similarity matrixes to spectral clustering. With texture feature, this paper carefully tuned the Gabor filter so as to generate a better texture. The analysis and segment result of these texture features by spectral clustering is presented in experiment section. With color feature, it discussed four color spaces and analyzed their segment result by spectral clustering. This paper also analyzed the impact of space feature to spectral clustering with our designed experiment. From those experiments we have summarized a list of conclusions. They are listed in the conclusion section.There are a lot of different similarity matrixes construction methods that have influence on spectral clustering. The analysis presented in this paper is not enough. There also many other aspects should be considered. We listed these future works in the conclusion section.