登入選單

返回Google圖書搜尋

A Continuation Method for Image Estimation and Segmentation

A. Rangarajan

出版	University of Maryland, Center for Automation Research, Computer Vision Laboratory, 1991
URL	http://books.google.com.hk/books?id=SZhNHAAACAAJ&hl=&source=gbs_api

註釋

Abstract: "The problems of image estimation and segmentation have been cast in a joint maximum a posteriori framework using Gibbs distributions defined over the intensities and line processes. MAP estimation is then reduced to minimizing an appropriate energy function defined on the intensity and line processes. The energy function typically has three components; (a) a measure of closeness to the data, (b) a weak constraint which assumes that the image is mostly smooth except at the discontinuities, and (c) penalties on broken contours, multiple edges, etc. In its most general form, the energy is highly non-convex causing deterministic relaxation techniques to converge to shallow, local minima