Distributed face detection system with complementary classifiers

Driven by the needs of various applications, we propose in this paper a distributed face finding system consisting of complementary classifiers. In this system, a grid pattern based face classifier utilizes low frequency information to quickly eliminate non face objects and inputs a small number of face candidate windows to subsequent computationally expensive face classifier(s). The subsequent classifiers use higher frequency information for refinement. Experimental results show that, on average, the distributed face detection system processes 8 images (768 x 512 pixels) per second with a high detection rate and very low false positives.