Rough Set Based FCM Algorithm for Image Segmentation

In this paper, a modified FCM (Fuzzy C-Means) algorithm based on Rough Set for image segment is proposed. The processes of the approach include two stages. In first stage, a cluster center set is built by reduction theory (the core of Rough Sets). In this stage, a decision table is designed firstly, where an initial cluster center set which contains granules with ill-defined boundaries is treated as an object, and each granule is treated as the object’s attribute. Discernibility of the initial cluster center sets in terms of attributes is then computed in the form of a discernibility matrix. Then using rough set theory, a number of decision rules are generated from the discernibility matrix. The rules represent rough clusters of points in the original feature space. In second stage, the eliminated cluster center set is input to FCM as the initial cluster center for the soft evaluation of the segmentation, where the fuzzy membership functions is modeled by a rule and utilized to compute the similarity of each image pixel. Experimental results demonstrate the efficiency and the effectiveness of the proposed method.