Robust self-weighting multi-view fuzzy clustering for image segmentation

Abstract

Traditional single-view clustering cannot process multi-view data well, so multi-view clustering is proposed. However, existing multi-view clustering algorithms have shortcomings in view information extraction, weight parameter selection, and noisy data processing. To overcome these drawbacks, we proposed a robust multi-view fuzzy clustering algorithm based on self-weighting that can handle noisy image segmentation. The algorithm first introduces the consistent fuzzy division matrix and balance coefficient into the general model to actively learn common and individual information among different views; secondly, it adopts a self-weighting strategy to judge the importance degree of each view for clustering; and finally, we construct fuzzy factors for each view based on the local space information and map the improved multi-view algorithm to the kernel space, thereby enhancing its robustness. In addition, we use an alternating optimization method to solve the optimization model and Zangwill’s theorem to prove the convergence of our algorithm. The algorithm has optimal evaluation metrics in the segmentation tasks of synthetic, natural, and remote sensing images, where the ACC values are all greater than 0.91 and the PSNR values are all greater than 20, giving superior segmentation capability and interference resistance compared to several comparative algorithms.