Comparative analysis of accuracy of fuzzy clustering methods applied for image processing

Abstract

Fuzzy C-Means (FCM) has been used in different aspects of machine learning, specifically in clustering and image processing. Several modified versions of FCM have been proposed to improve the accuracy. This paper also makes use of FCM in image processing using a new membership function applied in Bounded Fuzzy Possibilistic Method (BFPM) to provide a more flexible search space for data points with respect to all clusters. The method is proposed with regard to the challenges with conventional fuzzy methods. The paper evaluates the accuracy of BFPM in its partitioning strategy and membership assignments in compare with other advanced fuzzy, possibilistic, and other prototype-based partitioning methods. Promising results of tests performed on the benchmark image dataset Libras prove that BFPM performs better than other conventional methods.