Single Stage Face Detection Algorithm Based On Dual Shot

As a key technology in face information processing,face detection provides a theoretical basis for the application development of face alignment and face recognition,and has a wide range of application prospects in video surveillance and humancomputer interaction.Therefore,this paper proposes single stage face detection algorithm based on dual shot,and its main contents are as follows:Firstly,this paper introduces dual shot face detector based on improved SSD framework.The detector is mainly composed of feature enhance module,progressive anchor loss and improved anchor matching method.Feature enhance module is used to transfer the original feature maps to extend the single shot detector to dual shot detector.Progressive anchor loss computed by using two set of anchors is adopted to effectively facilitate the features.Improved anchor matching method integrates novel data augmentation techniques and anchor design strategies to provide better initialization for the regressor.Extensive experiments on popular benchmarks demonstrate the superiority of dual shot face detector.Then,this paper further optimizes algorithm based on the face detector,it mainly includes the following three items:（1）Equal-proportion interval anchor mechanism.It guarantees that different scales of anchor have the same density on the image,so that various scales face can approximately match the same number of anchors.（2）Scale compensation anchor matching strategy.When the anchor is matched,the face detector greatly increases the matched anchors of tiny and outer faces whose scales distribute away from anchor scales by reducing the IoU threshold,which notably improves the recall rate of these faces.（3）Max-out background label.In order to solve the problem that the false positive rate of shallow features is too high,this paper adopts the max-out classification strategy on the shallow feature detection layer conv3₃,Which can reduce the false positive rate of small-scale faces.Finally,the algorithm is validated on several public face detection databases.The experimental results show that the results of the algorithm in the FDDB,AFW and PASCAL FACE datasets are 98.26%,99.68%and 99.04%,respectively,WIDER FACE（easy:95.0%,medium:94.2%,hard:88.7%）.It is proved that the proposed algorithm has better performance and robustness for small-scale face and faces with large variations in occlusion,blur,illumination,pose.