The cluster is applied in image process for segment.K-means is populated for its simplicity and easily realization.This algorithm is liable to stuck at values which are not optimal and the result is relied on cluster center of initial selection.In order to overcome these drawbacks,a novel image segmentation algorithm (KFGA) is proposed.The first improvement is to Hybrid the genetic algorithm and K-means for searching the global optimum.The second improvement is to replace the Euclidean distance with feature divergence Inner product norm for increasing the Adaptability.The results of the experiment show that the algorithm has the better Adaptability and getting the correct global optimum.