The active contour models have been studied for image segmentation since late 1980s, and most of them find local minima in the corresponding energy function, therefore some recent works seek to compute global solutions. In this paper, we obtain a new fast global minimization algorithm by solving a recent convex image segmentation model through a gradient-based dual formulation of the minimization problem. The proposed method can achieve desirable segmentation with arbitrary initiation and avoids re-initializing. We demonstrate the efficiency of our method by testing it on images with additive Gaussian noise.
展开▼
