DISCOVERY AND LEARNING TECHNIQUES FOR PATTERN RECOGNITION.
Abstract
This paper describes a pattern recognition program which accepts any 20 x 20 black-and-white pattern, reduces it to a one-dimensional description of its contour, and identifies the pattern. Starting without operators or pre-set memory, the program learns via a feedback process, what characterizes each pattern type. It begins its identification of a new pattern by searching for matches between portions of the pattern's contour and learned stroke-like characteristics. A study of locations of certain of these matches, plus a comparison with memory, yields an identification of the pattern. The paper also discusses ideas for future pattern recognizers. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 18, 1964
- Accession Number
- AD0610725
Entities
People
- Leonard M. Uhr
- Rebecca C. Prather
Organizations
- System Development Corporation