Robust Facial Component Detection for Face Alignment Applications
Martin Urschler, Markus Storer, Horst Bischof, Josef Birchbauer · 2009
Biometric applications like face recognition/verification are an important topic in computer vision, both for research and commercial systems. Unfortunately, state of the art face recognition systems are not able to work with arbitrary input images taken under different imaging conditions or showing occlusions and/or variations in expression or pose. To support face recognition one has to perform a face image alignment (normalization) step that takes occlusions/variations into account. In this work we present a robust face normalization algorithm suitable for arbitrary input images containing a face. The algorithm is based on detecting face and facial component candidates and robustly voting for the best face and eyes. Our restrictions are a certain pose range (frontal to half profile) and suitable illumination conditions. Our algorithm is designed to deal with occlusions and its performance is shown on three publicly available image databases. 1