Fuzzy Image Segmentation Using Membership Connectedness

Fuzzy connectedness and fuzzy clustering are two well-known techniques for fuzzy image segmentation. The former considers the relation of pixels in the spatial space, but does not inherently utilize their feature information. On the other hand, the latter does not consider the spatial relations among pixels. In this paper, a new segmentation algorithm is proposed in which these methods are combined via a notion called membership connectedness. In this algorithm, two kinds of local spatial attractions are considered in the functional form of membership connectedness and the required seeds can be selected automatically. The performance of the proposed method is evaluated using a developed synthetic image dataset and both simulated and real brain magnetic resonance image (MRI) datasets. The evaluation demonstrates the strength of the proposed algorithm in segmentation of noisy images which plays an important role especially in medical image applications.