Optimization approaches to constraint satisfaction problems in computer vision

Abstract This paper describes several new image understanding methods based on parallel operation. There are several constraint satisfaction approaches using an energy minimization. We show how we reconstruct three-dimensional surfaces from contours without elevation data and sparse points of known elevation data using this approach. We also introduce Active Net using this approach, and apply this model to segmentation and binocular stereo matching. We experimented with these energy minimization approaches to solve the problems of early and intermediate levels of computer vision, and show some of the results of our recent research.