Date of Graduation
Statler College of Engineering and Mineral Resources
Lane Department of Computer Science and Electrical Engineering
Natalia A. Schmid.
As a successful technology in commercial as well as surveillance applications, face recognition has attracted significant attention. Wide range of applications using this technology has been a constant motivation for research developments over the past decade. Major issues with color images are illumination variation and pose change in which illumination variation has been overcome by involving Short and Long Wave Infrared imagery. However, properties of color images and Short and Long Wave Infrared images are different. Their crossmatching presents a great challenge.;In this thesis, we propose a methodology that will be able to crossmatch color face images and Short Wave Infrared face images reliably and accurately. We first adopt a recently designed Boosted Local Gabor Pattern Improved (LGPI) encoding and matching technique to encode face images in both vsible and Short Wave Infrared (SWIR) spectral bands. We then apply feature selection methods to prune irrelevant information in encoded data and to improve performance of the Boosted LGPI technique. In this thesis, we propose three novel feature selection methods: (1) Genuine segment score thresholding, (2) d' -based thresholding and (3) two Adaboost inspired methods. We further compare the performance of the original Boosted LGPI face recognition method with the method involving feature selection step. Under a general parameter set up, a significant performance improvement is observed and perfect verification performance is achieved.
Boothapati, Sirisha, "Feature Selection Methods for Boosted Crosspectral Face Recognition" (2011). Graduate Theses, Dissertations, and Problem Reports. 3304.