GPU accelerated face detection

Recently many-core graphic processor units (GPUs) are delivering impressive power for general purpose computing applications. Thanks to their high memory bandwidth and computing throughput, GPUs could often significantly accelerate many applications. In this paper, we present a CPU-GPU cooperative implementation for a Viola-Jones [5] based face detection system. The experiment results show that our face detector running on a GTX280 graphics card could achieve an over 20X speed-up compared with the CPU equivalent on an Intel core 2 duo processor, while maintaining exactly the same detection quality. We also show that our implementation offers good scalability in terms of image size.