In view of the existence of obvious over-segmentation problem in traditional watershed algorithm, an improved color image segmentation algorithm is proposed in this paper. This algorithm combines clustering and improved watershed algorithm methods. In HSV color space, the algorithm firstly utilizes clustering to classify pixels with similar features, then applies seed region growing algorithm for the regions generated by watershed segmentation algorithm. Furthermore, region merging approach is used to merge remaining small regions. Experimental results show that this algorithm reduces over-segmentation phenomenon significantly, and improves image accuracy while maintaining strong robustness and adaptability.