Alignment by Maximization of Mutual Information

  • Authors:
  • Paul Viola;William M. Wells, III

  • Affiliations:
  • Massachusetts Institute of Technology, Artificial Intelligence Laboratory, 545 Technology Square, Cambridge, MA 02139. E-mail: viola@ai.mit.edu;Massachusetts Institute of Technology, Artificial

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 1997

Quantified Score

Hi-index 0.00

Visualization

Abstract

A new information-theoretic approach is presented for finding thepose of an object in an image. The technique does not requireinformation about the surface properties of the object, besides itsshape, and is robust with respect to variations of illumination. Inour derivation few assumptions are made about the nature of theimaging process. As a result the algorithms are quite general and mayforeseeably be used in a wide variety of imagingsituations.Experiments are presented that demonstrate the approachregistering magnetic resonance (MR) images, aligning a complex 3Dobject model to real scenes including clutter and occlusion, trackinga human head in a video sequence and aligning a view-based 2D objectmodel to real images.The method is based on a formulation of the mutual informationbetween the model and the image. As applied here the technique isintensity-based, rather than feature-based. It works well in domainswhere edge or gradient-magnitude based methods have difficulty, yetit is more robust than traditional correlation. Additionally, it hasan efficient implementation that is based on stochasticapproximation.