Improving indoor and outdoor face recognition using unified subspace analysis and Gabor features

Lighting variation is one of the major problems for face recognition. Most of the current face recognition studies on lighting problem are based on the face image set taken under controlled laboratory lighting or normal indoor lighting. In the recent FRVT 2002 test, it is found that the best face recognition systems are not sensitive to normal indoor lighting changes, but have a significant drop in performance on the outdoor probe face image set. This clearly indicates that recognition of faces in outdoor images needs to be a focus of future research. In this paper, we address the lighting variation problem using several simple and practical techniques. Using the unified subspace method in combination with wavelet local features and appropriate training data selection, we improve the indoor and outdoor face recognition significantly.