Abstract: Recently,Laplacian eigenmaps has been used in the color image segmentation based on cluster algorithm as an improving means．The graph weight matrix which is learned from Gaussian method may not characterize the locally geometric structure of the data points．As a result this method can’t tell the fuzzy edges sufficiently．To overcome this problem,the locally linear embedding in kernel space algorithm is proposed．Firstly,the simple pixel’s eight-nearest neighbor method is introduced to get weighted graph．Secondly,With the kernel trick,Laplacian reconstruction coefficients used to reflect the similarity between data points in the high dimension space can be achieved,as well as the subspace of feature vector．Finally,Fuzzy C-means clustering algorithm is used in the subspace to distribute labels for each pixel and achieve the segmentation of color images．The result of experiments compared with the results of the Laplacian eigenmaps segmentation method provided shows the encouraging improvements of the new algorithm.

Key words: LE,FCM,LLE,Kernel space,Color image segmentation