Superpixels-based Robust Fuzzy C-means Clustering Algorithm with Boundary Retune for Image Segmentation

In this paper, a superpixels-based robust algorithm for image segmentation based on fuzzy C-means (FCM) is investigated, which performs image segmentation with the help of global information and local one (SGFCML). Superpixels can remain the similar intensity information in each superpixel, and reduce the noise influence, which is a popular preprocessing method. In this paper, after the superpixel preprocessing, in the first stage, for the initial segmentation an improved fuzzy local information c-means algorithm is performed, which introduces coefficient of variation of local windows and pixel gray value similarity simultaneously. Because these two factors are introduced, the trade-off between smoothing and clustering can be controlled more accurately and the robustness can be enhanced. In the second stage, local information is utilized to perform segmentation in the regions near the initial boundary, and weighted voting strategy applied to refine the boundary. Experimental results indicate that the proposed algorithm is effective over conventional FCM and other extended FCM algorithms.