Abstract: Based on the analysis of advantages and disadvantages of the self-adaptive FCM image segmentation algorithm, an improved self-adaptive fuzzy C-means clustering algorithm is proposed.By construct a two-dimension histogram with pixel neighborhood space information, as a cluster sample for self-adaptive FCM, the dimension of sample space is reduced, and then, the problem of self-adaptive FCM on low rate of convergence and sensitive to noise is solved.At last, the improved algorithm is improved to have higher rate of convergence, higher accuracy of separation, and more robustness to noise, by contrast to self-adaptive FCM algorithm on image segmentation to a noise contained image.