This paper proposes an optimized face recognition system for embedded hardware environment. A novel active infrared lighting mechanism is designed to improve quality of face image. Adaboost algorithm is used to locate regions of interest in face image. Based on the guidance of heuristic rules, an improved DLA algorithm is designed to extract feature in frequency domain. A PNN-based classifier is implemented to achieve recognition result. The system architecture is also discussed based on these key methods. We realize the system on ARM11 CPU and the test result shows high recognition accuracy with high running speed.