Segmentation by Fusion of Self-Adaptive SFCM Cluster in Multi-Color Space Components

This paper proposes a new, simple, and efficient segmentation approach that could find diverseapplications in pattern recognition as well as in computer vision, particularly in color imagesegmentation. First, we choose the best segmentation components among six different colorspaces. Then, Histogram and SFCM techniques are applied for initialization of segmentation.Finally, we fuse the segmentation results and merge similar regions. Extensive experiments havebeen taken on Berkeley image database by using the proposed algorithm. The results show that,compared with some classical segmentation algorithms, such as Mean-Shift, FCR and CTM, etc,our method could yield reasonably good or better image partitioning, which illustrates practicalvalue of the method.