ON THE POSSIBILITY OF AUTOMATICALLY RECOGNIZING CERTAIN VISUAL PATTERNS (VOZMOZHNOSTI AVTOMATICHESKOGO RASPOZNAVANIYA NEKOTORYKH VIZUALNYKH OBRAZOV),
Abstract
The report states that the recognition of such visual recognition objects as human faces can play some role in problems of automatic pattern recognition, in spite of the apparent complexity of the recognition of faces as compared, for example, with recognition of standard letters or numbers. Some of the recognition features common to faces and to printed characters are first discussed. The results are reported of an experiment in which 60 full-face female and 140 male photographs were superimposed on one another. Each negative was exposed for (1/k)-th part of the normal exposure (k is the number of superimposed pictures, and ranged from 1 or 2 to a maximum). The results showed unexpectedly that composites of male photographs could be clearly distinguished from composites of female photographs when other outward attributes (long hair, moustache, etc.) were eliminated. It is concluded that male and female faces constitute separate compact sets. Some statistical aspects of the results are briefly discussed. This may serve as a starting point for ascertaining the possibility of automatic differentiation between male and female faces. The size of the appropriate sample is briefly treated and it was found that about 20 photographs sufficed to obtain the 'average' characteristics. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 11, 1967
- Accession Number
- AD0670899
Entities
People
- G. S. Ramishvili
Organizations
- National Air and Space Intelligence Center