Face Detect Method Based On Fully-connected Neural Network

With the rapid development of computer vision in deep learning,the technology of face detection is also in a high-speed development.The task of face detection is to find all existing faces in the input image.However,the traditional face detection algorithms need to face a variety of facial expressions,gestures,lighting,scene,scale and other challenges.Inspired by the traditional Adaboost detection algorithm,researchers propose a large number of algorithms based on deep convolution of the neural network.From the training multiple the face classifier network within sliding window to the end-to-end training of the face detector algorithm.In this paper,we use the latest general object detection algorithm to train the face detection model of the full convolution network.Base on this work,we the improvement the model for face detection,and multi-scale position sensitive face detection(FaceRFCN)is proposed,The classification algorithm is combined with the low-level and high-level feature map of the network,and the multi-scale RoI is used to improve the algorithm recall.The model proposed in this paper has high precision and high speed.In several face detection brenchmark,it reachs the performance of the world’s most advanced face detection algorithm.