Face Detection Research Based On Adversarial Deep Learning

In recent years,deep convolutional neural networks(CNN)have significantly improved the accuracy of image classification and object detection.Compared with image classification,object detection is a more challenging task,because object detection requires the precise location of objects and becomes more complex,so more complicated methods are needed to solve,and the comparison of the detection of object in complex situations difficult.Relying on the region proposal algorithm to predict the object's location,the object detection network development pushes the object detection to a new height.The research direction of this paper is to apply the method of object detection and generative adversarial network combination to the detection of face in occlusion.The main research contents and results of this article mainly include:Firstly,based on Fast R-CNN object detection method,the method of selective search is used to extract the UMDfaces from the face dataset,and Fast R-CNN is applied to the face detection.In addition,the set combines Faster R-CNN object detection method to achieve face detection.Secondly,in the generative adversarial network based on Fast R-CNN experiment,combining Fast R-CNN with the generative adversarial network to achieve face detection under occlusion conditions,inspired by the independent training of Faster RCNN,RPN(Region Proposal Network)replaces selective search method to extract face image region proposals,and combined with the generative adversarial network based on Fast R-CNN for training together.In the experiment process,the accuracy of the detection of face using the RPN network and the generative adversarial network based on Fast R-CNN of confrontation network is 88.71%.Finally,we combining Faster R-CNN and generative adversarial network,the end-to-end training generative adversarial network based on Faster R-CNN was implemented.The final accuracy of face occlusion detection was 89.34%.This experimental result is better than the former two methods.The experimental results show that the combination of generative adversarial network and Faster R-CNN has certain feasibility for the detection of face under occlusion conditions.