Uncooled Infrared Imaging Face Recognition Using Kernel-Based Feature Vector Selection
Abstract
A considerable amount of research has been recently conducted on face recognition tasks, due to increasing demands for security and authentication applications. Recent technological developments in uncooled IR imagery technology have boosted IR face recognition research applications. Our study is part of an on-going research initiated at the Naval Postgraduate School that considers an uncooled low-resolution and low-cost IR camera used for face recognition applications. This work investigates a recent approach which approximates nonlinear kernel-based methods at a significantly reduced computational cost. Our research was applied to an IR database. Results show that this scheme may perform sufficiently close to its "kernelized" version considered in a previous study, at a fraction of the computational cost, provided that the associated parameters are well tuned. The thesis considers a relative comparison between the two algorithms, based on identification and verification experiments and considers a statistical test to investigate whether classification performance differences may be considered statistically significant. Results show that, from a cost perspective, a low-resolution uncooled IR camera in conjunction with a low computational-cost classification scheme can be embedded in a robust face recognition system to efficiently address the issue of authentication in security-related tasks.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2006
- Accession Number
- ADA456994
Entities
People
- Ioannis M. Alexandropoulos
Organizations
- Naval Postgraduate School