MACHINE LEARNING PROCESSES APPLIED TO PATTERN RECOGNITION
Abstract
The theory of the principal methods for pattern recognition is summarized. Emphasis is on the division of the whole process into three phases: calibration from a sample set of patterns, derivation of numerical properties, and a decision process for determining the class of a pattern. Several mathematical processes are suggested for finding useful properties, including equations which define invariants to acceptable distortions. It is shown that if an efficient and consistant decision process is used, then the property search is easier since the addition of useless or redundant properties cannot degrade the performance. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 14, 1961
- Accession Number
- AD0282549
Entities
People
- Arthur E. Laemmel
Organizations
- New York University Tandon School of Engineering