A novel non-convex regularization method for image segmentation
Zhilong Zhao, Yu Du Han, Hui Wang, Fengqi Yu · 2011
This paper proposes a new variational method with a non-convex regularization term which is introduced to restore high quality image. Non-convex regularization has advantages over convex regularization such as total variation (TV) for image segmentation. In practical, the used of the non-convex regularization is limited by the difficulty of the minimization. Through the variation splitting technology, we develop a new fast minimization algorithm to solve the non-convex problem for image segmentation. The new algorithm has higher efficiency and more robust to the choice of parameters. Experimental results illustrate the performance improvements by using our method.