Multilevel Relaxation in Low Level Computer Vision.
Abstract
Much work in low level computer vision has involved the dense interpolation or approximation of sparsely-known or noisy data. A few examples are image smoothing, surface interpolation, and optic flow computation. A recent approach to these problems has formulated them in terms of optimization or constrained minimization. In general these techniques are equivalent to solving elliptic partial differential equations with boundary conditions and constraints. In either formulation, these problems can be solved by a class of algorithms well suited to computer vision. Variational (cost minimization) and local constraint approaches are generally applicable to problems in low-level vision (e.g., computation of intrinsic images). Iterative relaxation algorithms are natural choices for implementation because they can be executed on highly parallel and locally connected processors. They may, however, require a very large number of iterations to attain convergence. Multi-level relaxation techniques converge much faster and are well suited to processing in cones or pyramids. These techniques are applied to the problem of computing optic flow from dynamic images.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1982
- Accession Number
- ADA124054
Entities
People
- Frank Glazer
Organizations
- University of Massachusetts Amherst