Head pose estimation with improved random regression forests

Abstract

Head pose estimation is an important step in many face related applications. In this paper, we propose to use random regression forests to estimate head poses in 2D face images. Given a 2D face image, Gabor filters are first applied to extract raw high-dimensional features. Linear discriminant analysis (LDA) is then used to reduce the feature dimension. Random regression forests are constructed in the low dimensional feature space. Unlike traditional random forests, when generating tree predictors in the forests we weight the features according to the eigenvalues associated with their corresponding LDA axes. The proposed method has been evaluated on a set of 2D face images synthesized from the BU-3DFE database and on the CMU-PIE database. The experimental results demonstrate the effectiveness of the proposed method. © Springer International Publishing 2013.