On the Use of Distance and Information Measures in Pattern Recognition and Applications.

Abstract

In the development of pattern recognition theory during the past few years, there has been active interest in comparing the relative merits of various distance and information measures by evaluating the error bounds. Normally the tighter the error bounds a distance or information measure can provide, the better features it can select. There are now at least fifteen different measures available. They have two main advantages. (1) They are generally easier to calculate than the error probability itself which normally cannot be computed exactly. These measures can provide good upper and/or lower bounds of the error probability. (2) Features can be selected and ordered to maximize (or in some cases minimize) the measures in order to minimize indirectly the error probability. Linear transformation for dimensionality reduction can also be derived from these measures. This paper is concerned with the theoretical problems and the applications of these measures.

Document Details

Document Type
Technical Report
Publication Date
Aug 01, 1975
Accession Number
ADA015799

Entities

People

  • Chia‐Hung Chen

Organizations

  • University of Massachusetts Dartmouth

Tags

DTIC Thesaurus Topics

  • Dimensionality Reduction
  • Pattern Recognition
  • Probability
  • Recognition
  • Signal Processing

Readers

  • Computational Modeling and Simulation
  • Computer Vision.
  • Operations Research

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms