Research on Far-Field Face Detection for Recognition

Far-field face detection is the first step of automatic face recognition in video surveillance. In this paper, we propose a framework on extraction of high-resolution images in video that subjects are far from camera. To guarantee the tradeoff between accuracy and speed, our method uses four techniques including motion detection using Gaussian Mixture Models (GMMs), skin model detection and Adaboost-based detector with the expansion of skin region. Moreover, face hallucination based on eigentransformation is adopted to obtain high-resolution images for recognition. Experimental results on real data show that our system can acquire acceptable detection rate with promising speed.