C-means clustering fuzzified by two membership relative entropy functions approach incorporating local data information for noisy image segmentation

Abstract

In this paper, C-means algorithm is fuzzified and regularized by incorporating both local data and membership information. The local membership information is incorporated via two membership relative entropy (MRE) functions. These MRE functions measure the information proximity of the membership function of each pixel to the membership average in the immediate spatial neighborhood. Then minimizing these MRE functions pushes the membership function of a pixel toward its average in the pixel vicinity. The resulting algorithm is called the Local Membership Relative Entropy based FCM (LMREFCM). The local data information is incorporated into the LMREFCM algorithm by adding to the standard distance a weighted distance computed from the locally smoothed data. The final resulting algorithm, called the Local Data and Membership Relative Entropy based FCM (LDMREFCM), assigns a pixel to the cluster more likely existing in its immediate neighborhoods. This provides noise immunity and results in clustered images with piecewise homogeneous regions. Simulation results of segmentation of synthetic and real-world noisy images are presented to compare the performance of the proposed LMREFCM and LDMREFCM algorithms with several FCM-related algorithms.