Object Oriented Segmentation of Images.

Abstract

The central theme of this research project is recovery of object shapes from noisy images. The main mathematical techniques are energy functionals and gradient descent. The goal is to construct vertically integrated models capable of incorporating constraints imposed by various objectives such as noise suppression, boundary detection, shape description and object recognition. The difficulty with this approach is that these functionals are very hard to implement on account of their nonlinerity and the need to find discontinuous solutions. Therefore, the main emphasis of this research project has been on the formulation of approximate models. The key ingredient is representation of the discontinuity locus by a continuous variable which is then used to control processes at hand such as noise suppression or stereo matching. Specific formulations are developed for nonlinear smoothing of images, stereo matching which takes into account half-occlusions, recovery of surface shapes from shading data in the presence of noise and discontinuities and finally, curve evolution for locating and smoothing object boundaries. (MM)

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 21, 1994
Accession Number
ADA290792

Entities

People

  • Jayant Shah

Organizations

  • Northeastern University

Tags

Communities of Interest

  • Energy and Power Technologies
  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Algorithms
  • Artificial Intelligence
  • Boundaries
  • Computer Science
  • Computer Vision
  • Detection
  • Differential Equations
  • Discontinuities
  • Equations
  • Image Processing
  • Image Recognition
  • Object Recognition
  • Pattern Recognition
  • Personal Information Managers
  • Recognition
  • Recovery
  • Vascular System Injuries

Readers

  • Acoustics.
  • Calculus or Mathematical Analysis
  • Computer Vision.