Modified Possibilistic Fuzzy C-means Clustering Driven by Weighted Residuals for Image Segmentation

Abstract To address the shortcomings of the probability fuzzy C-means algorithm (PFCM) in image segmentation—specifically its sensitivity to noises and the occasional formation of coincident clusters, a modified probability fuzzy C-means algorithm driven by weighted residuals (WRMPFCM) has been proposed. First, a modified possibilistic fuzzy C-means clustering algorithm (MPFCM) has been developed by adjusting the penalty term in the objective function of PFCM to mitigate the issue of coincident clusters. Second, residuals are introduced into the objective function to estimate noise, and l_2-norm of weighted residuals is employed to constrain model training, which effectively reduces noise interference. Finally, spatial neighborhood information is incorporated to account for the spatial relationships and similarities between pixels in the image, thereby enhancing segmentation accuracy. Experiments conducted on synthetic images, medical images, and real-world images demonstrate that WRMPFCM exhibits significant noise resistance, improved segmentation accuracy, and enhanced adaptability across various image segmentation scenarios.