Robust Cluster Center Localization Rough Clustering Algorithm for Image Segmentation

This study presents a robust initial class center localization algorithm for rough clustering image segmentation, which aims to improve segmentation accuracy based on the location of target areas within images. First, a global separation threshold for image features is calculated based on Sobel edge detection results. Next, a new kernel function is designed to compute the boundary weights of image pixels, generating a boundary image. Finally, a competitive mechanism is proposed to generate connected regions from the boundary image, with the average feature values of the top C largest connected regions serving as initial class centers. The rough clustering algorithm is then utilized to achieve image segmentation. Experimental results demonstrate that this method significantly enhances target localization precision, exhibits good stability, and achieves higher segmentation accuracy.