A convolutional neural network combined with aggregate channel feature for face detection

Face detection has been studied intensively over the past several decades and achieved great improvements via convolutional neural network (CNN) which has greatly improved the performance in image classification and object detection. In this paper, similar to the idea of R-CNN [1], we present a new method that combines the aggregate channel features (ACF) [2] and CNN for face detection. The proposed method uses ACF to select the possible human face regions and then trains a CNN model to filter out non-face candidates. Then we merge the results of ACF and CNN to get the final detection window(s). Evaluations on two popular face detection benchmark datasets show that our method outperforms the ACF method and has achieved competitive performance against the state-of-the-art algorithms.