Human Image Understanding
Abstract
The goal of the effort is to develop and empirically evaluate a theory (Recognition-by-Components (RBC)) of real-time human target identification which assumes that objects are represented as an arrangement of simple generalized-cone volumes. The fundamental assumption of RBC is that a particular set of these convex components, called geons, can be derived from invariant properties of edges in a 2-D image. If an arrangement of three geons can be recovered from the input, objects can be quickly recognized even when they are occluded, rotated in depth, novel, extensively degraded, or embedded in a scene. The report describes the research on consequences of various forms of image degradation, the exploration of the role of surface features, the attentional demands of object recognition, formal modeling of object recognition, and extensions to scene perception and extensions to scene perception and expert identification. Keywords: Pattern recognition; Perception; Vision; Image understanding.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1989
- Accession Number
- ADA204490
Entities
People
- Irving Biederman
Organizations
- University at Buffalo