Fast face detection on mobile devices by leveraging global and local facial characteristics

Face detection is an extensively studied problem in the last few decades. Recently, significant improvements have been achieved by the deep neural network (DNN) solution. It is however challenging to apply the DNN technique to mobile devices directly due to its limited computational power and memory. In this work, we present a proposal generation acceleration framework for real-time face detection. Specifically, we adopt the convolutional neural network cascade as the baseline and develop an acceleration scheme to speed up its inference time. The acceleration scheme is motivated by the observation that the computational bottleneck of the baseline arises in the proposal generation stage, where each level of the dense image pyramid has to go through the network. To address this issue, we reduce the number of image pyramid levels by utilizing both global and local facial characteristics (i.e., global face and facial parts). We conduct experiments on public WIDER-face and FDDB datasets and show that the proposed scheme offers the state-of-the-art face detection performance at a much faster speed.