Interactive Image Segmentation Based on Graph Cut
Minyong Shi · Jisuanji fangzhen · 2008
A novel Graph Cut based algorithm for interactive segmentation was proposed. Firstly, the initial curve obtained by Graph Cut was fit with Non-Uniform Rational B-Spline (NURBS). Due to the local controllability of NURBS, the control points were considered as the states of Hidden Markov Model (HMM), and the boundary features and uniformity along the boundary were integrated as the observations. Then, a state transition model based on contour smoothness constraint was calculated. Finally, the optimal contour could be achieved efficiently by Viterbi algorithm. Experimental results show the robustness, easiness and efficiency of the method.