FDA based fast haar-like feature selection for cascaded AdaBoost face detection

Viola´s framework of cascaded AdaBoost classifiers is one of the best approaches for real-time face detection. However, training of cascaded AdaBoost classifiers is time-consuming, needs days or even weeks. A FDA based fast haar feature selection method is proposed in this paper, which use statistics of training samples. Time complexity of our method is O(N+T), comparing to O(NTlog(N)) given by Viola´s original method. We also present a method based on l_o normalized FDA, which gives a faster detector together with fast training.