On Statistical and Structural Feature Extraction.

Abstract

The problem of extracting effective features has played a major role in pattern recognition studies. Most effort, however, has been made on the selection of statistical (or mathematical) features. Examples of these features are the orthogonal transforms of vector measurements. Although such features may not have physical meaning, they are usually effective and computable. The main drawback with the statistical feature extraction is that it is difficult to take contextual dependence into consideration. In many patterns such as the imagery data, there is a close interrelation among various parts of a pattern. Such interrelation representing the structural properties must be taken into account to achieve the best recognition performance. In this paper we consider the structural feature extraction in a more general sense by taking into account the relationships among parts of a pattern, and the physical, perceptual, psychological, and/or physiological factors which may be translated from description to mathematical expressions suitable for automatic recognition. The paper reviews the problems in both statistical and structural feature extraction, compares the feasible approaches and examines the statistical and structural mixed model for feature extraction. (Author)

Document Details

Document Type
Technical Report
Publication Date
May 19, 1976
Accession Number
ADA028405

Entities

People

  • Chia‐Hung Chen

Organizations

  • University of Massachusetts Dartmouth

Tags

DTIC Thesaurus Topics

  • Automatic
  • Extraction
  • Feature Extraction
  • Measurement
  • Pattern Recognition
  • Recognition
  • Structural Properties

Readers

  • Computer Vision.
  • Systems Analysis and Design

Technology Areas

  • AI & ML