ON THE LINEAR AND QUADRATIC DISCRIMINATORS FOR PATTERN RECOGNITION.
Abstract
In this report, two problems in pattern recognition are considered: (1) when the likelihood ratio criterion results in a quadratic discriminant function, it may be possible to factor the quadratic form into two linear discriminant functions. It is shown that the probabilities of misclassifications are the same for both the linear discriminant functions and that both of them should be used for classification. (2) When the likelihood criterion results in a linear discriminant function, it is shown that the estimate of the 'squared distance' has a positive bias and hence the probability of misclassification is underestimated. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 21, 1967
- Accession Number
- AD0659296
Entities
People
- Poduri S. R. S. Rao