Enhancing interdisciplinary image segmentation through a Gaussian-based modified local consensus spatial fuzzy approach.

This study aims to introduce a generic fuzzy-based approach tailored explicitly for classifying images originating from an array of diverse sources, having varying degrees of spectral and spatial resolutions, inhomogeneity, artifacts, and entirely distinct features. The proposed Gaussian-based Modified Local Consensus Spatial Fuzzy (GMLCSF) approach stands out as an innovative solution differing from the traditional fuzzy-based approaches and the advanced methods in the domain, if multiple imaging sources and artifacts with uncertainties are present in the datasets, i.e. satellite images and medical images, where classified visual data is essential. The initial kick of the proposed approach comes from the histogram peak associative rule, which identifies the number of clusters and initializes the centers intelligently. The consensus-inspired local spatial membership function is incorporated with the standard global membership function to eliminate the noise and inhomogeneities, during the estimation of belongingness to a class. The Gaussian, geometric, and local consensus-based spatial information is formulated to elevate the efficacy and accuracy of the framework irrespective of image sources and uncertainties. The proposed GMLCSF is an iterative process, hence to decide the stopping criteria, we have considered three conditions and discussed them in the proposed method section in detail. The proposed framework is developed and simulated in MATLAB and tested on remote sensing and MRI datasets. The quantitative effectiveness of the GMLCSF over state-of-the-art techniques has been estimated by partition coefficient, entropy, and spectral angle distance. The qualitative results as classified images were analyzed in detail and again the superiority of the approach over state-of-the-art techniques in the domain has been observed.