Relative entropy fuzzy c-means clustering

Pattern recognition is a collection of computer techniques to classify various observations into different clusters of similar attributes in either supervised or unsupervised manner. Application of fuzzy logic to unsupervised classification or clustering methods has resulted in many wildly used techniques such as fuzzy c-means (FCM) method. However, when the observations are too noisy, the performance of such methods might be reduced. Thus, in this paper, a new fuzzy clustering method based on FCM is presented and the relative entropy is added to its objective function as a regularization function to maximize the dissimilarity between clusters. Several examples are provided to examine the performance of the proposed clustering method. The obtained results show that the proposed method has a very good ability in detecting noises and assignment of suitable membership degrees to the observations.