Geometric Invariants in Object Recognition.
Abstract
The project has involved the use of invariants in computer vision and has shown that invariants can greatly improve the efficiency of object recognition. As part of the project, two kinds of invariants have been studied. (1) Geometric invariants, which assume that the geometry of the object is given. The objective here is to use descriptors of the object that are independent of the geometric transformations that complicate object recognition, such as change of the viewpoint from which the object is seen, or small deformations. (2) Physical invariants, which take into account the physical process by which the image is obtained such as irradiation by visible light, infra-red, radar, sonar etc. We borrow methods from physics to apply invariants similar to energy and momentum of the physical process. In these processes there are usually more unknowns than equations. The invariants help in putting additional constraints on the underdetermined set of equations.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 31, 1995
- Accession Number
- ADA311344
Entities
People
- Isaac Weiss
Organizations
- University of Maryland