Variational problems have proved of value in many image processing and analysis applications. However increase of sensor resolution as occurred in medical imaging and experimental fluid dynamics demand for adapted solving strategies to handle the huge amount of data. In this paper we address the decomposition of the general class of quadratic variational problems, which includes several important problems, such as motion estimation and image denoising. The basic strategy is to subdivide the originally intractable problem into a set of smaller convex quadratic problems. Particular care is taken to avoid ill-conditioned sub-problems. We demonstrate the approach by means of two relevant variational problems.
展开▼
