I. INTRODUCTION
Accurate facial feature extraction is important for face
alignment, which is an indispensable processing step
between face detection and recognition. This paper is to
build a feature-extraction system that can be used for face
recognition in embedded and/or consumer applications.
This imposes specific requirements to the algorithm in
addition to extraction accuracy, such as real-time
performance under varying imaging conditions and
robustness with low-cost imaging hardware.