Color image segmentation combining mean shift andweighted spectral clustering
融合均值漂移和加权谱聚类的彩色图像分割

This paper proposed a novel algorithm for improving efficiency of color image segmentation. Firstly, it filtered noise and transformed input image from pixel-based to region-based model by using mean shift algorithm, and composed the input image after mean shift procedure by some disjoint regions. Then it treated the regions as nodes in image plane and applied a graph structure to represent them. Finally it applied the weighted spectral clustering algorithm which merged the information of area differences and spatial distances among the regions to perform final clustering, and abtained the result of image segmentation. Experimental results show that the proposed algorithm is effective on color image segmentation, and it also has the properties of low computation cost, keeping boundary well and reducing noise interference.