Face template creation: Is Centroid Method a suitable approach?

The paper presents a research and development of a face template creation sub-system, which serves a part of a face-based identification system. This system is intended to be used in surveillance camera systems. The paper addresses a problem of contradicting requirements on the face recognition system - near-real time recognition and a low error rate. The research, we have conducted, has shown that a solution may be found in designing an adequate face template database and in applying appropriate face template creation algorithms. Face templates have a significant influence on the correct classification rate, while the number of temples is directly proportional to the classification time. The results of our research have so far proved that using the Centroid Method for face templates creation is the best performing approach. The paper presents two novel methods for a face template creation in order to investigate whether the Centroid Method may be outperformed. The performance of proposed face template methods is evaluated on images from an extensive IFaViD database assembled during our research, images were captured by a surveillance camera system.