Fast Image Segmentation Using Variational Optimization Methods With Edge Detector

Abstract

In this work, we apply techniques in variational optimization to image segmentation. We study three different segmentation models: one is based on the active contour method, the second is based on a piecewise constant level set method, and the last uses a continuous max-flow min-cut model. We obtain significantly better segmentation results in the first and the third model by including an experimental edge detector. The first model is a special case of the minimal partition problem, the second model uses discontinuities of piecewise constant level set functions to represent interfaces between the region of interest and the background, and the third model uses a spatially continuous max-flow min-cut framework which is a very efficient method to segment images. The first two models are non-convex and may contain many local solutions, but the last model is a convex optimization problem and therefore finds the global solution.