RPCCL Clustering and Its Evaluation in Image Segmentation

In this paper, we propose a new image segmentation approach based on rival penalized controlled competitive learning (RPCCL) and color quantization technique. First, we perform median filtering on input image. Second, the initial color centers are selected by color quantization algorithm. Then after several iterations, the RPCCL clustering converges and produces the final segmentation results. We carry out experiments and quantitative evaluation based on Berkeley Segmentation Database (BSD300). The results show that RPCCL method is superior to K-means clustering.