Non-linear Boosting Regression for Multi-view Face Alignment

Abstract

Face alignment is a critical problem in many face related applications such as facial expression analysis,face recognition,etc.In this paper,we use local textures of each label point to predict the displacement of each label point by applying nonlinear boosting regression based on Haar rectangle feature,and develop a novel real-time multi-view face alignment system based on the active shape model.Experiments on two independent datasets show that our algorithm is much faster,more accurate and robust than the classic active shape model and outperforms recently improved algorithm,too.It has significant practical value in real applications.