A new face database simultaneously acquired in visible, near infrared and thermal spectrum
In this paper we present a new database acquired with three different sensors (visible, near infrared and thermal) under different illumination conditions. This database consists of 41 people acquired in four different acquisition sessions, five images per session and three different illumination conditions. The total amount of pictures is 7.380 pictures. Experimental results are obtained through single sensor experiments as well as the combination of two and three sensors under different illumination conditions (natural, infrared and artificial illumination). We have found that the three spectral bands studied contribute in a nearly equal proportion to a combined system. Experimental results show a significant improvement combining the three spectrums, even when using a simple classifier and feature extractor. In six of the nine scenarios studied we obtained identification rates higher or equal to 98%, when using a trained combination rule, and two cases of nine when using a fixed rule.