Performance Characterization of Image Stabilization Algorithms.

Abstract

This paper compares three image stabilization algorithms when used as preprocessors for a target tracking application. These algorithms vary in computational complexity, accuracy, and ability. Algorithm 1 is capable of only pixel-level realignment of imagery, while Algorithms 2 and 3 are capable of full subpixel stabilization with respect to translation, rotation, and scale. The algorithms are evaluated on their performance in the stabilization of one synthetic forward looking infrared (FLIR) data set and two real FLIR imagery data sets. The evaluation tools incorporated include mean square error of the output data set and the overall performance of an automatic target acquisition system (developed at the Army Research Laboratory) that uses the algorithms as a front end preprocessor. We find that for this tracking application, extremely accurate subpixel stabilization is a requirement for proper operation. We also find that in this application, Algorithm 3 performs significantly better than the other two algorithms.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Apr 01, 1996
Accession Number
ADA313877

Entities

People

  • Rama Chellappa
  • Stephen B. Balakirsky

Organizations

  • University of Maryland

Tags

DTIC Thesaurus Topics

  • Accuracy
  • Acquisition
  • Algorithms
  • Automatic
  • Computational Complexity
  • Data Sets
  • Errors
  • Military Research
  • Rotation
  • Target Acquisition
  • Target Tracking
  • Test And Evaluation
  • Translations

Fields of Study

  • Computer science

Readers

  • Image Processing and Computer Vision.
  • Military and Counterinsurgency Studies.
  • Neural Network Machine Learning.