A hybrid fuzzy dissimilarity histogram and automatic fuzzy clustering framework for robust color image segmentation

One limitation of clustering algorithms is the manual specification of the number of clusters. Density Peak (DP) clustering can automatically determine the optimal number, but in image segmentation, it may face memory overflow due to large similarity matrices from medium-sized images. this paper introduces the Fuzzy Dissimilarity Histogram (FDH) method, to address this drawback. FDH enhances image contrast using fuzzy background information. FDH improves noise robustness while preserving image details. The image histogram is computed and combined with spatial features to boost segmentation accuracy. This combined histogram better distinguishes different image regions, allowing enhanced identification of specific feature areas. Smoothing the histogram and removing irrelevant data improves processing speed and accuracy. The proposed method applies an Automatic Fuzzy Clustering Framework (AFCF) for image segmentation. It integrates the superpixel concept into the DP algorithm to improve computational efficiency. Fully automated clustering is achieved by initially applying a density equilibrium algorithm followed by fuzzy c-means clustering based on prior entropy, enhancing segmentation quality. Experiments on synthetic and real images demonstrate that the proposed approach outperforms advanced clustering algorithms in segmentation quality and processing time for color images. Additionally, this method significantly improves segmentation results for Multi-scale Morphological Gradient Reconstruction (MMGR)-AFCF, as well as Simple Linear Iterative Clustering (SLIC)-AFCF and Linear Spectral Clustering (LSC)-AFCF algorithms.