Two-Step Segmentation for Speedup of Convergence via Preprocessing

This paper introduces an integrated two-steps segmentation algorithm in the framework of Mumford-Shah functional. Note that the efficiency and convergence speed of the active contour-based segmentation algorithms are strongly dependent of selections of initial curves. Therefore a preprocessing step is introduced and integrated to construct an initial level set which is very closer to the boundaries of objects. As a result, a fast convergence speed is achieved. Furthermore the algorithm has better flexibility on segmentation of different kinds of images when some preprocessing techniques like denoising, edges enhance are used. In addition, the local minimal problem in the classical algorithm also can be eliminated or improved by choosing better diffusion approaches. The resulting algorithm has also been demonstrated by several cases.