Segmentation Model Based On The Global Convex Two-phase Image Segmentation

Image segmentation is an important application aspect of image processing technology. The goal of it is to separate the image into regions with different characteristics through some kind of operation, extract the objects interested and display it. In recent years, the geometric active contour model in image segmentation has emerged many results in the field of two-phase image segmentation based on variational level set method. However, with the image segmentation in various fields getting more and more widely used, the traditional limitations of image segmentation methods has seriously affected the image segmentation results. Traditional image segmentation model has only local energy minimum for it is not convex. Different initial positions of level set may lead to different energy minimums of different points, and thus get different segmentation results. Another drawback of such model is that the level set function evolution equation requires repeated iteration; the lower computational efficiency limits the requests of real-time applications with computing requirements. In this paper, we mainly studied two-phase piecewise constant images and two-phase piecewise smooth image segmentation deeply, mainly including the following aspects:First, as the results of image segmentation depends a lot on the initial location of level set, based on a new piecewise smooth global convex segmentation model proposed by Bresson, we proposed an improved one; second, systematically studied the application of the dual method in image processing, Bresson applied the dual method to two-phase piecewise constant image and the two-phase piecewise smooth image segmentation, this article created the noise segmentation model of two-phase piecewise constant images depending on the regional distribution of the image noise based on GCS, respectively were used to separate the noise in line with Gauss distribution and Rayleigh distribution; third, introduced the application of Split Bregman iterative method in image processing, and expanded Split Bregman iterative method from two-phase piecewise constant image to two-phase piecewise smooth image segmentation; Furthermore, all the models mentioned above are implemented, and has been used in artificial images, remote sensing images and real medical image segmentation. Many experiments denote that we get the satisfying results using the corresponding image segmentation models. Lastly, future research directions are put forward according to the problems and limitation occurred during my research work.