CLASS MEMBERSHIP CRITERIA AND PATTERN RECOGNITION.

Abstract

The object of the work presented here is to provide improved conceptual and computational tools for handling the class membership problem: to decide whether an observation consisting of numerical values assigned to each of N parameters is or is not due to some phenomenon known descriptively in terms of a sufficiently large set of such observations, or to which of several phenomena the observation should be assigned. To each data vector in N space, representing a single observation, are annexed the elements of higher order moment tensors constructed from it to form an 'extended vector.' From the extended vectors associated with a cluster of data points is constructed an 'extended covariance matrix' which serves to define distance from the cluster and in a fashion which reflects the shape of the given cluster. Distance of a new observation (data point) from a given cluster, as measured in this metric, provides a means for deciding whether the observation belongs to the cluster. A pattern, consisting of a large number of points in N space, is represented by its 'extended covariance matrix.' The latter can be regarded as a vector in 'pattern space' and hence amenable to the same considerations employed in the class assignment problem. (Author)

Document Details

Document Type
Technical Report
Publication Date
Apr 01, 1968
Accession Number
AD0834208

Entities

People

  • R. P. Eddy

Tags

DTIC Thesaurus Topics

  • Covariance
  • Data Science
  • Identification
  • Information Science
  • Observation
  • Pattern Recognition
  • Recognition

Fields of Study

  • Mathematics

Readers

  • Computer Vision.
  • Linear Algebra
  • Systems Analysis and Design

Technology Areas

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