Abstract: Graph cuts is an interactive segmentation algorithm based on boundary and region properties of objects in images. The region term in conventional graph cuts is based on Gaussian Mixture Model(GMM). However, it is not only a slow process, but sometimes it can’t describe the distribution of pixels in objects precisely. This paper proposes an improved algorithm based on K-means clustering graph cuts. Its evaluation is performed using both phantoms and real Magnetic Resonance Imaging(MRI) of brain, the effectiveness and efficiency of the proposed algorithm are showed. And in particular, an accurate and robust results in segmenting images with noise and intensity non-uniformity with a low computational cost can be achieved.
Magnetic Resonance Imaging(MRI) of brain