Baselining illumination variables for improved facial recognition system performance

Lighting has been shown to be one of the largest degradation factors of facial recognition systems. Several model and algorithmic approaches have been attempted to reduce this problem with varying success. We present here results from an experimental program which quantifies lighting variables including direction, temperature and colour, intensity, diffuseness and glare, as tested on the face-it and SQIS facial recognition systems. We conclude by recommending the use of lighting of colour temperature 4000 K and diffuse lighting from two horizontally aligned sources, 508 lx at 45° to the forward gaze line of the subject.