 |
Technical Program
Paper Detail
| Paper: | MA-P4.10 |
| Session: | Image Registration/Alignment and Mosaicking |
| Time: | Monday, October 9, 09:40 - 12:20 |
| Presentation: |
Poster
|
| Title: |
SIMPLER ALTERNATIVES TO INFORMATION THEORETIC SIMILARITY METRICS FOR MULTIMODAL IMAGE ALIGNMENT |
| Authors: |
Shannon Hughes; Princeton University | | | | | Ingrid Daubechies; Princeton University | | |
| Abstract: |
Although mutual information (MI) methods are widely used in image registration, applications would benefit from a computational simplification. While stochastic in concept, MI methods actually seek the best alignment by maximizing a similarity metric that is deterministically computed from the two images. We study this metric and find that maximizing it is equivalent to minimizing a distance metric between certain equivalence classes of images. We then create new metrics that preserve only this aspect of the MI metric and observe that they experimentally attain equal alignment accuracy while significantly decreasing computation. We conclude that this property of MI alone may suffice for accurate registration. |
|
