LEARNING CONTROL SYSTEMS AND PATTERN RECOGNITION.
Abstract
Topics covered include finite state sequential machines, thin film deposition research, construction of chip matrix receptors, neural networks of the retina, phototransistor response testing, position and velocity detection systems, and character recognition. A feasibility study of the theory and design of position and velocity detecting systems using pattern recognition concepts is presented. Design parameters of the system's receptor are considered. The results indicate that the required size of the receptor matrix is relatively large for available solid state receptors. Methods of receptor resolution improvement through input signal perturbation are presented. The results indicate that an order of magnitude improvement in position detection accuracy can be obtained by appropriately choosing the objects size, sensor element geometry, and the amplitude of the perturbation signal. A character recognition machine that is insensitive to translation and, to a lesser degree, dilations and angular orientation of the input samples is described. The system consists of three stages: (1) a receptor to make certain measurements on the input patterns to be classified; (2) a preprocessor to subdivide the pattern set into sixteen subsets; and (3) a categorizer to separate the members of the individual subsets. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1969
- Accession Number
- AD0684325
Entities
People
- B. H. Gilpin
- E. J. White
- E. S. Mcvey
- J. W. Moore
- P. F. Chen
Organizations
- University of Virginia