RegionBoost Learning for 2D+3D based Face Recognition

This paper describes an improved boosting algorithm, named RegionBoost, and its application in developing a fast and robust invariant Local Binary Pattern histogram based face recognition system. We propose to use a multi-classifier where each classifier, an AdaBoost of feed-forward back-propagation network, is trained using a single Sub-Window of the whole image, the classifiers are finally combined using the “Sum Rule”. Only the best matchers, selected by running the Sequential Forward Floating Selection (SFFS), are exploited in the fusion step. In our opinion our method (based on local AdaBoost) partially solves the problem of redundancy among global AdaBoost selected features, with a manageable computational requirement. Finally, we propose a systematic framework for fusing 2D and 3D face recognition sytems.