Image Matching by Maximisation of Mutual Information

We have applied the method designed by Paul Viola, from M.I.T. This method, originally for finding the pose of an object in an image, has been extended to the case of 2D image matching.

The method tries to find the transformation that best matches two images, in optimising the mutual information between the two images. Intuitively, if the two images are correctly matched, knowing one image gives information about the other. Therefore their mutual information is high. Conversely, two independent signals or images will have a very low mutual information.
 

  • Entropy and Mutual Information
  • Method
  • Optimisation Scheme
  • Results
    • You can download the technical report [693 Kb, *.ps.gz] describing and experimenting the method. I wrote this technical report as part of my first-year report. 
    Sebastien Gilles, now at IMEDIA Research Group, Paris, France.