In this paper, the main application image processing, manifold learning and the method of Gaussian mixture model for dimensionality reduction and cluster analysis, the image color information are all studied. First, the color data access algorithm is introduced, secondly, the manifold learning in the local linear embedding (LLE) algorithm is used in color analysis; then the results of an evaluation criteria, LLE parameters in the criteria automatic selection algorithm are presented; meanwhile, the result of the operation of the different color space LLE are also tested and analyzed. Finally, the application of a Greedy EM-based Gaussian mixture model for improving the operation of the HSI space under LLE results of experiments is analyzed. It indicates that the algorithm can automatically determine the number of clusters, and achieve a better clustering result.