%0 Journal Article
%T Visible and Infrared Face Identification via Sparse Representation
%A Pierre Buyssens
%A Marinette Revenu
%J ISRN Machine Vision
%D 2013
%R 10.1155/2013/579126
%X We present a facial recognition technique based on facial sparse representation. A dictionary is learned from data, and patches extracted from a face are decomposed in a sparse manner onto this dictionary. We particularly focus on the design of dictionaries that play a crucial role in the final identification rates. Applied to various databases and modalities, we show that this approach gives interesting performances. We propose also a score fusion framework that allows quantifying the saliency classifiers outputs and merging them according to these saliencies. 1. Introduction Face recognition is a topic which has been of increasing interest during the last two decades due to a vast number of possible applications: biometrics, video surveillance, advanced HMI, or image/video indexation. Although considerable progress has been made in this domain, especially with the development of powerful methods (such as the Eigenfaces or the Elastic Bunch Graph Matching methods), automatic face recognition is not enough accurate in uncontrolled environments for a large use. Many factors can degrade the performances of facial biometric system: illumination variation creates artificial shadows, changing locally the appearance of the face; head poses modify the distance between localized features; facial expression introduces global changes; artefacts wearing, such as glasses or scarf, may hide parts of the face. For the particular case of illumination, a lot of work has been done on the preprocessing step of the images to reduce the effect of the illumination on the face. Another approach is to use other imagery such as infrared, which has been showed to be a promising alternative. An infrared capture of a face is nearly invariant to illumination changes and allows a system to process in all the illumination conditions, including total darkness like night. While visual cameras measure the electromagnetic energy in the visible spectrum (0.4每0.7ˋ米m), sensors in the IR respond to thermal radiation in the infrared spectrum (0.7每14.0ˋ米m). The infrared spectrum can mainly be divided into reflected IR (Figure 1(b)) and emissive IR (Figure 1(c)). Reflected IR contains near infrared (NIR) (0.7每0.9ˋ米m) and short-wave infrared (SWIR) (0.9每2.4ˋ米m). The thermal IR band is associated with thermal radiation emitted by the objects. It contains the midwave infrared (MWIR) (3.0每5.0ˋ米m) and long-wave infrared (LWIR) (8.0每14.0ˋ米m). Although the reflected IR is by far the most studied, we use thermal long-wave IR in this study. Figure 1: A face captured under (a) visible spectrum, (b)
%U http://www.hindawi.com/journals/isrn.machine.vision/2013/579126/