Investigations of linear classifiers applied to uncooled infrared imaging face recognition

Infrared imaging devices offer the main advantage of being robust to ambient illumination changes and recent technological advances have significantly improved the resolution of cheaper uncooled infrared (IR) versions to the point where they may become widely applicable in the IR face recognition field. This study applies two linear classification schemes; the principal component analysis and the Fisherface procedure to the recognition of faces collected using an uncooled IR camera under indoor controlled conditions. Results show the selected uncooled IR camera has sufficient resolution to allow for discrimination between the subjects contained in our experimental database collected. Results show also consistent better performances for LDA than for PCA-based schemes. We also discuss the impact of the first few top eigenvectors on PCA-based classification performances.