December 5, 1995
This talk will address the problem of recovering structured visual information from imperfect images. Several examples of this problem will be given. These examples arise from the need to develop enabling technologies for use in automotive driving assistance, aircraft landing guidance, highway traffic maintainence, and image database query systems.
The problem is tackled under a Bayesian paradigm that involves
deformable template shape models, Markov random field feature models,
and multi-resolution optimization algorithms. Extensive
experimental results that illustrate the performance of
the method will be presented.