On Non-Parametric Discrimination Using Series Expansions and Projection Pursuit.

Abstract

This article develops a rigorous non parametric theory of optimal binary discrimination using series expansions. A relationship is presented between minimum scatter and limiting nearest neighbor error rate and its pertinence of optimal discrimination. A general consistency result for series expansions is then given. This motivates a data driven consistent projection pursuit alogorithm for constructing an orthonormal basis for discriminant expansions. It is seen for projection pursuit discrimination that limiting nearest neightbor error rate plays an important role as mean squared residual error and relative entropy do in projection pursuit regression and density estimation, respectively. Finally an application to motion detection of optical point sources is given and a numerical experiment is carried out.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Feb 26, 1987
Accession Number
ADA178455

Entities

People

  • Lee K. Jones

Organizations

  • The Catholic University of America

Tags

Communities of Interest

  • C4I

DTIC Thesaurus Topics

  • Algorithms
  • Classification
  • Computer Programs
  • Consistency
  • Data Sets
  • Delta Functions
  • Detection
  • Discrimination
  • Estimators
  • Fourier Series
  • Hilbert Space
  • Military Research
  • Optical Detection
  • Probability
  • Random Variables
  • Sequences
  • Standards

Readers

  • Calculus or Mathematical Analysis
  • Regression Analysis.
  • Vision Science/Vision Psychology/Cognitive Neuroscience.