Simple head pose estimation in still images

Robust, reliable head pose estimation is a key step in many practical applications involving face analysis tasks. We address the problem of head pose estimation in still gray scale images, assuming a standard camera with limited resolution details. To achieve the proposed goal, we rely on the standard pattern recognition approach: we describe the previously detected faces with easy-to-compute image features (such as integral image projections, Local Binary Pattern - LBP and Histogram of oriented Gradient - HoG), that are subsequently feed into a machine learning system (namely a Multi-Layer Perceptron - MLP) that will approximate the head angle. We have thoroughly evaluated our system on the HPEG and Columbia publicly available databases.