The Scaling Problem in Visual Pattern Recognition.
Abstract
The human pattern recognition system has been modeled as a system utilizing a low-pass, spatially-filtered, Fourier transform to represent input patterns and stored prototypes. This model is used to infer the existence of possible mechanisms underlying the scale invariant aspect of the human pattern recognition system. Two hypotheses are suggested: scaling (size normalization) of input patterns, or the storage of multiple-prototypes based on size. Experiments are carried out to distinguish between the two mechanisms. It is concluded that both mechanisms are utilized at different levels within the visual process. A revised model is developed to adequately account for the experimental results. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1976
- Accession Number
- ADA035290
Entities
People
- Douglas D. Carpenter
Organizations
- Air Force Institute of Technology