Fast and Global Minimization Convex Multiphase Active Contours
Lifen Zhou, Shunji Zhang · 2016
Biomedical images usually have multiple regions of interest, which lead to the study of multi object segmentation in this area.Multiphase level set active contour model using a gradient descent method to minimize energy non convex, so not only will be the local minimum, but also lead to erroneous segmentation, the fast algorithm can not be carried out, can not be solved quickly.Aiming at the above problems,we propose a multi object image segmentation algorithm of global convex coupling.First, in order to avoid the segmented regions overlap and vacuum, we put two-phase level set Chan Vese model was extended to four phase, and the introduction of edge stopping function improved regularization process, and then based on this model of convex optimization, to obtain the global optimal solution, finally, using dual minimization method for the fast calculation.Experimental results show that the proposed model can not only obtain the global optimal solution, but also greatly improve the efficiency and accuracy of the segmentation.