Peripheral nerve segmentation based on the improved Grab Cut

Peripheral nerve segmentation is difficult because of similar characteristics. In this paper, an interactive segmentation method based on K-Harmonic Means clustering and improved Grab Cut is proposed. Firstly, peripheral nerve images are processed with K-Harmonic Means clustering algorithm, and images are divided into some regions where the pixel characteristics are similar. Then the Gaussian mixture model(GMM) parameters are initialized for every region through K-Harmonic Means clustering. Finally, the parameters are estimated with the iteration method to minimize the energy function and achieve correct segmentation results. Experimental results show that the proposed method is effective for peripheral nerve segmentation and has achieved good performance.