Cascade Stacked Hourglass Model For Facial Key-points Detection

The face recognition process can be divided into face detection,facial key-points detection and facial feature matching.First,face detection is the process of determining the number of faces from an image and determining the location of faces.Second,facial keypoints detection is the process of determining facial specific information based on face detection.Last,feature matching compare features extracted from the image with features in the datasets.Because of facial key-points detection can be affected by background,lighting,face pose,face expression,and decoration,resulting in insufficient accuracy of detection.In view of the above-mentioned problems,the paper conducts an in-depth study of facial key-points detection,and cascading for facial key-points detection,that is,the process is divided into two steps of coarse positioning and fine positioning.Compared with the known facial keypoints detection's methods,the main innovations of the paper are: 1)The stacked hourglass network is used for the coarse location of the facial key-points detection;2)The face divide into seven parts during the fine location,(In other papers,the face divide into six parts),feature extraction is performed separately,and then facial key-points are corrected.The current process is more precise than the previous method.Experiments on LFPW and Helen datasets show that the cascade method and the network structure used in the paper can effectively reduce the error rate of keypoint detection.