Abstract:

This paper presents progress toward an integrated face detection and demographic analysis system that is robust and works in real-time. Faces are detected and extracted using the very fast algorithm recently proposed by Viola & Jones. Detected faces are passed to a novel demographics classifier which uses the same architecture as the face detector. This demographic classifier is extremely fast, yet delivers error rates slightly better than the best known classifiers. Demographics information, since it can be noisy in realistic situations, is integrated across time for each individual. The final demographic classification combines the estimates from many facial detections in order to significantly reduce error rate. The entire process runs faster than 10 fps on an 800 MHz Intel PIII.