University of Houston
Department of Computer Science

In partial fulfillment of the Requirements for the Degree of
Master of Science

Pradeep Buddharaju
will defend his thesis

FACE RECOGNITION IN THE THERMAL INFRARED SPECTRUM

Abstract
This thesis presents a statistical face recognition method based on the thermal infrared portion of the electro-magnetic spectrum. In this region images are formed primarily due to emission.herefore, they do not depend on the existence and intensity of an external light source. They are also less dependent on the incident angle. In terms of algorithmic approach we have adopted a spectral analysis methodology. In contrast to geometric approaches spectral analysis is more resilient to stochastic facial changes as well to disguise attempts. Our algorithm proceeds in four (4)stages: In the first stage we segment the imagery using adaptive fuzzy connectedness segmentation to remove the background. In the second stage we divide the segmented face into its spectral components using a bank of K Gabor filters. We associate K Bessel probabilistic models to these filtered images to obtain 2K Bessel parameters, which constitute the feature vector. In the third step, we short-list high probability subjects by applying the L^2-metric on the Bessel feature vectors. In the fourth and final step, we feed the short-listed set to a Bayesian classifier to find the exact match. Our method introduces two important novelties: the adaptive fuzzy connectedness segmentation and the Bayesian classification steps. In the experiments with the Equinox thermal face database, our method performed better than competing approaches with accuracy scores exceeding 90%. This work is sponsored by an NSF research grant on information assurance.

Date: 30, NOVEMBER, 2005
Time: 02:30 PM
Place: 550-PGH

Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Ioannis Pavlidis