The Research And Application Of Spectral Clustering Algorithm Based On Density Estimation

Spectral clustering algorithm is a clustering algorithm developed in spectral graph partitioning theory, which has been widely concerned. Be compared to other clustering algorithms, spectral clustering algorithm not only has a solid theoretical foundation, but also be able to cluster in arbitrary shape of sample space, and can convergence to the global optimum,so it is very popular in recent years. Similarity matrix is an extremely important part in spectral clustering algorithm. Therefore, this paper explores how to construct a similarity matrix, and proposes two methods based on density estimation to construct similarity matrix,and applies these two algorithms to color image segmentation. The main work of this paper is as follows:To begin with, a spectral clustering algorithm based on local density and geodesic distance is proposed.Firstly, the local density of each point is calculated to find the nearest high density point, and choose marginal points and non-marginal points. Then, the edge between the marginal point and its nearest high density point is created, and edges between the k nearest neighbor points and non-marginal points are created as well. Finally, the geodesic distance and similarity can be computed to clustering. The experiments results show that this algorithm can obtain higher accuracy in dealing with the adhesive datasets.Then, a spectral clustering algorithm based on density coefficient and shared nearest neighbor is proposed. Firstly, the density coefficient of each point is computed, and the adaptive scale parameter is calculated according to the density coefficient. Secondly, the number of the nearest neighbor points and similarity are calculated. Finally, the points are clustered by the similarity. The experimental results show that the method of calculating the similarity by using the density coefficient can better describe the similarity between the points with complex distribution, and get a better result.At last, the two kinds of spectral clustering algorithm based on density estimation are applied to color image segmentation. Due to the color image with a large number of pixels and clustering of pixels requires more resources to be consumed. In order to easily use the spectral clustering algorithm for image segmentation,the SLIC algorithm is used to per segment and generate a number of super pixels. Then the LUV color feature of each pixel is extracted, and the color features are clustered by two kinds of spectral clustering algorithm based on density estimation, which can realize the color image segmentation. The experimental results show that the two proposed algorithms can be successfully applied to image segmentation, and have achieved good results.