A Local Zernike Moment-based Fuzzy C-means Algorithm for Segmentation of Brain Magnetic Resonance Images

Segmentation of brain magnetic resonance (MR) images is essential for computer-aided diagnosis. However, accurate segmentation is mainly affected due to the presence of noise artifact in MR images. In this paper, to reduce the effect of noise, we compute the similarity weights between the pixel values in the moment domain using Zernike moments (ZMs) as opposed to the existing approaches which use spatial domain techniques for this purpose. The ZMs used in the proposed method uses low and middle order ZM coefficients to compute similarity weights for providing spatial information. The higher order ZM coefficients that are susceptible to image noise are discarded. The proposed method is observed to be very effective for brain MR image segmentation. Detailed experimental results are performed on simulated and clinical datasets.