Architecture for Higher Level Digital Image Processing.

Abstract

This report analyzes the segmentation properties of the Peleg relaxation scheme for the one-dimensional, two-label case. It is shown that if the probability of either label is identically zero, then that label probability will remain identically zero for all iterations. However, if the label probabilities are non-zero, then the label probability at each iteration will either increase toward unity or decrease toward zero, depending on whether the average probability of the neighborhood is above or below a threshold determined by the relationship of the compatibility coefficients. Further analysis of the boundaries between regions of different labels shows that boundary stability requires that the net effective coefficients for both labels be equal. In addition, it is shown that the speed at which ambiguity is resolved is maximized if non-zero compatibility coefficients are allowed for alike labels. There has been no indication that increased speed alters the results. Extensions to multiple-label cases are discussed. (Author)

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 31, 1979
Accession Number
ADA092153

Entities

People

  • Arden Helland

Organizations

  • University of Maryland

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Ambiguity
  • Artificial Intelligence
  • Boundaries
  • Coefficients
  • Computer Programs
  • Computer Science
  • Computer Vision
  • Computers
  • Digital Image Processing
  • Digital Images
  • False Alarms
  • Image Processing
  • Iterations
  • Probability
  • Universities

Readers

  • Calculus or Mathematical Analysis
  • Computer Vision.