View-based recognition under illumination changes using local features
Christian Wallraven, HH Bülthoff · MPG.PuRe (Max Planck Society) · 2001
We present a view-based face recognition system which combines elements from both feature-based and appearance-based approaches to increase recognition performance under illumination changes. It uses corners and their local neighborhood at several scales to construct local features which form the representation of each face image. Matching feature sets takes into account both configurational and appearance-based similarity. We present recognition results on a highly realistic synthetic face-database demonstrating the system's ability to tolerate illumination changes. In addition, the proposed framework agrees well with current findings from psychophysics.