Face detection under variable lighting based on resample by face relighting

Different environment illuminations have a great impact on face detection. In this paper, we present a solution based on face relighting technology. The basic idea is that there exists nine harmonic images that can be derived from a 3D model of a face, and by which we can estimate the illumination coefficient of any face samples. Using an illumination radio image, we can produce images under new lighting conditions. To detect faces under certain lighting condition, we relight original face samples to get more new faces under kinds of possible lighting condition and add them to the training set. Our experimental results on Support Vector Machine (SVM) turn out that the relighting subspace is effective on detection under variations of the lighting conditions. Moreover, if we relight original face samples to new samples under different illuminations, the collected example sets will be multiplied. We use the expanded database to train an AdaBoost-based face detector and test it on the MIT+CMU frontal face test set. The experimental results show that the data collection can be efficiently speeded up by the proposed methods. The later experiment also verifies the generalization capability of the proposed method.