Color image segmentation using mean shift and improved spectral clustering

A novel approach is presented for color image segmentation. By incorporating the advantages of mean shift (MS) segmentation and spectral clustering (SC) method, the proposed approach provides effective and robust segmentation. Firstly, input image is transformed from a pixel-based to a region-based model by using the MS algorithm. The input image after MS segmentation is composed of multiple disjoint regions that preserve the desirable discontinuity characteristics of the image. Then the regions are treated as nodes in the image plane and a graph structure is applied to represent them. The final step is to apply the improved SC to perform globally optimal clustering. To avoid some incorrect partitioning when considering each region as one graph node, we assign different numbers of nodes to represent the regions according to area ratios among the regions. In addition, K-harmonic means (KHM) instead of K-means is applied in the improved SC procedure in order to enhance its stability and performance. The superiority of the proposed approach is demonstrated and examined through a mass of experiments using color images.