Application of Improved Fuzzy C-Means Algorithm Based on Mahalanobis Distance in Image Segmentation

Traditional fuzzy C-means exhibits poor performance on images when dealing with images possessing complex textures or similar grayscale levels. To address the limitation, an enhanced fuzzy C-means based on Mahalanobis distance is proposed in this paper, which significantly improves both accuracy and robustness when applied to image segmentation tasks. Specifically, the algorithm accurately measures the similarity between pixels by incorporating Mahalanobis distance, taking into account the covariance between pixels. The distance metric is computed using the \(l_{2,p}\) -norm, which considers the distance between pixels in different feature spaces. Moreover, by extracting multi-dimensional features, the algorithm comprehensively describes the visual and statistical characteristics of the images. Additionally, the efficient utilization of local information and the optimization of variables to control cluster size significantly enhance the algorithm’s efficiency. Experimental results on various image types validate the significant improvements in image segmentation quality achieved by the proposed method.