Self-sparse fuzzy clustering with automatic region merging approach for image segmentation

Abstract

Image segmentation based on fuzzy clustering has made enormous progress in recent years. However, the performance of existing methods degrades in noisy conditions and suffers from losses of foregrounds and the emergence of backgrounds. To settle these issues, in this paper, two methods – proposed method 1 and proposed method 2 – based on self-sparse fuzzy clustering algorithm (SSFCA) are proposed. The proposed methods use a normal shrink (NS) denoising algorithm as a pre-processing step for the suppression of noise. The output of NS undergoes the segmentation block SSFCA to get the segmentation output. The segmentation output is then subjected to a post-processing block for the removal of isolated pixels. In proposed method 1, the morphological cleaning operation acts as a post-processing block, whereas it is the morphological cleaning operation along with the automatic region merging approach in proposed method 2. The automatic region merging is achieved using connected component filtering based on the area density balance strategy (CCF-ADB). The proposed methods are tested on different datasets, and experimental results demonstrate their superior performance and effectiveness as compared to the state-of-art algorithms.