Image Understanding by Image-Seeking Adaptive Networks (ISAN).
Abstract
A remarkably simple, experimentally inspired, new theory of vision os presented. The theory takes into account the parallel architecture, the adaptive phenomena and the efferent control system which have been demonstrated in the vision systems of organisms. Further the complexities of visual receptive fields are made use of to explain the speed, noise resistance, consistencies and holistic aspects of perception. In this theory image understanding is achieved by image seeking adaptive networks that differentially amplify images of interest without first breaking them down into elementary components. A computer implementation of the theory demonstrates that the mechanisms postulated are feasible. A number of experiments with the model address critical aspects of image understanding and demonstrate that images of interest are captured reliably even in large amounts of noise, or in spite of position and/or size changes. Subjective edges, and other Gestalt like images, i.e. horizon and terrain are also seen by ISAN's basic network. Some implications for general vision are outlined.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 10, 1987
- Accession Number
- ADA186214
Entities
People
- D. N. Spinelli
Organizations
- University of Massachusetts Amherst