Semi-supervised Robust NRFCM for Image Segmentation with Pairwise Constraints

Clustering algorithms are increasingly employed for the image segmentation. By incorporating the spatial information and the term used in the punishing the distance, a new robust fuzzy c-means (NRFCM) algorithm was proposed for effectively improving the quality of the image segmentation. Its main characteristics are as follows:(1) The negative influence of the noise can be effectively reduced by using a penalty on the distance between one sample and clusters, (2)The segmentation of noises in images can be also avoided by bringing in the cluster weight. However, this algorithm cannot effectively use those given supervised information. So, in this paper, we propose here an effective semi-supervised robust NRFCM for image segmentation with pair-wise constraints (semi-NRFCM). Experiments show that the newly developed algorithm can effectively improve the quality of the image segmentation. Further, the experiments show that Semi-NRFCM is more suitable for the image segmentation with noises when comparing with NRFCM, FASTFCM and FCMS_1.