Person identification by face recognition on portable device for teaching-aid system: Preliminary report

We propose a face recognition system to identify a person and obtain his/her information, especially for teaching-aid contexts. This system is based on the communication between a portable device and a server. We evaluate face detection-recognition methods provided by OpenCV that will be used in the system. We also combine these methods with our illumination normalization and prove it can improve the detection and the recognition rate. With haar-based face detection and the illumination normalization, detection rate is stable at 95% in simple and severe illumination situations. Using Fisherface method with normalization, three training images per person are enough to achieve on average 96.4% recognition rate on Yale B Extended Database. Online prototype has been built and achieves up to 10 fps in performance.