Hybrid Image Segmentation Using RPCCL Clustering and Region Merging

Abstract

Image segmentation is very important to image analysis and satisfying methods are still unfounded. In this paper, we propose a new hybrid segmentation approach based on rival penalized controlled competitive learning (RPCCL) and region merging scheme. In the first, we performed median filtering on input image, and then selected initial color centers by using color quantization technique. During the RPCCL clustering, we merged some close centers to reduce classes. In the end, small regions were merged to produce the final segmentation results. Compared to original RPCCL, our method can overcome over-segmentation and obtain better results.