Robust K-means based Active Contours for Fast Inhomogeneity Image Segmentation

A novel robust K-means based active contours model is proposed to segment medical images with various noise and intensity inhomogeneities. Relying on the correntropy-based image features, the model uses the local adaptive weights to be robust to various noises. Moreover, the combination of information in the global and the local regions ensures that our approach is extremely hard to trap into a local minimum. To avoid the re-initialization and shorten the computational time, we use the signed distance functions to regularize the level set functions, and adopt the iteratively re-weighted method to accelerate our algorithm during the contour evolution. Experimental results show that our algorithm can fast achieve the robust segmentation results in the presence of the intensity inhomogeneities, various noise and blur.