Clustering based Image Segmentation via Weighted Fusion of Non-local and Local Information

In this paper, we introduce a novel and effective clustering model combining the non-local and local information for the image segmentation. Recently, the non-local information has attracted much attention in the area of image processing for its excellent ability to handle the noise. Specifically, in this new model, we do an automatically weighted fusion of the non-local and local information of the image in the objective function of K-means. Thus, in the smoothing areas, the non-local information reduce the impact of noise in the region; and in the edge of regions, the local information help to keep the image details. The proposed model is a general optimization problem which can be solved by the iterative refinement technique like fuzzy c-means or K-means, and it can automatically balance the contribution of non-local and local information. Verified by the experimental results on image segmentation, the proposed model is effective to improve the performance of clustering.