1 August 2007 Uncooled infrared-imaging face recognition using kernel-based generalized discriminant analysis
Author Affiliations +
Abstract
We investigate the application of the kernel-based generalized discriminant analysis (GDA) approach to the recognition of infrared face images collected using a low-resolution uncooled IR camera. Results show that a low-cost, low-resolution IR camera combined with an efficient classifier is a usable tool in uncooled IR face recognition applications, and that best GDA-based recognition performance improves over that obtained with the fisherface approach by 3.96 percentage points, from 94.59% to 98.55%, on the data considered. This study also investigates the effects the number of projection vectors used in the GDA step, the kernel expression, and the specific distance type have on recognition performance. Results show the Mahalanobis angular distance to be the best choice, and that the recognizer computational load may be reduced by decreasing the number of eigenvectors selected in the GDA projection step without significant impact on recognition performance.
© (2007) Society of Photo-Optical Instrumentation Engineers (SPIE)
Dimitrios I. Domboulas, Monique P. Fargues, Gamani Karunasiri, "Uncooled infrared-imaging face recognition using kernel-based generalized discriminant analysis," Optical Engineering 46(8), 087201 (1 August 2007). https://doi.org/10.1117/1.2769638 . Submission:
JOURNAL ARTICLE
6 PAGES


SHARE
Back to Top