Selection of discriminating characteristics by partial similarity for the two class recognition problem of artificial intelligence
Institution: | Oregon State University |
---|---|
Department: | Mathematics |
Degree: | MS |
Year: | 1971 |
Keywords: | Artificial intelligence – Computer programs |
Record ID: | 1489247 |
Full text PDF: | http://hdl.handle.net/1957/45551 |
I have developed what I believe to be a very general method of feature extraction. The key to the generality of the method is that we examine each individual characteristic separately with a minimum of dependence upon other characteristics. The basic idea is to discover discriminating characteristics by a linear investigation of each sample in each class of objects being examined for differences. This is facilitated by what I term "partial similarity," that is, we look at only those samples which are partially similar to a base sample with respect to previously discovered discriminating characteristics. The method was investigated by the use of a FORTRAN program using hand printed capital A's and R's as the two classes to discriminate between.