A Region Generation based Model for Occluded Face Detection
Jin Qi, Chong Mu, Ling Tian, Fankai Ran · Procedia Computer Science · 2020
Face detection is a necessary first-step in face recognition systems and other applications, such as intelligent interaction, identity verification, and mobile social networks. However, there is a lack of considering face occlusion in existing detection methods, resulting in the difficulty using these methods in practical scenarios. In this paper, we propose a novel detection model named FSG-FD with attention mechanism, and aim to detect face with occlusion. Experiments on several datasets (including occlusion and mutilscale features of faces) show that our proposed algorithm achieves 9.4% improvement than faster RCNN on wider dataset and 85.1% average precision on self-labeled monitoring dataset.