Pattern-based feature set for efficient segmentation of color images using modified FCM clustering

Based on the color feature of the image pixels, color image segmentation assumes that; distinct clusters of homogenous colors in the image correspond to significant objects in the image. Therefore, each cluster designates a group of pixels with comparable color characteristics. The present research work proposes a novel Modified kernel-based Fuzzy C Means Clustering (MKFCMC) method for color image segmentation using three stages: Pre-processing, Feature extraction, and Segmentation. In the pre-processing stage, the input image is filtered using Weiner Filtering model. The next stage is feature extraction in which shape index histogram-based features, improved local gradient pattern-based features, and color features are extracted. Finally, segmentation is done by the Modified Kernel Fuzzy C means (MKFCM) algorithm. In this MKFCM-based segmentation process, the optimal centroid selection is carried out using optimization algorithm named Self Improved Snake Optimization algorithm. Finally, a performance comparison is made between the proposed MKFCMC model and the standard state-of-the-art models in terms of accuracy, specificity, sensitivity, F1-score and other metrics, thereby establishing the superiority of proposed method.