Multi-view Face Detection And Facial Landmark Detection

With the development of computer vision technology,face recognition has been widely used in many real applications.Face recognition generally consists of face detection,face alignment and facial feature match.Our research work focused on face detection and face alignment.Details are as follows:1.This paper proposed a combined method to merge multiple face bounding boxes generated in face detection by Real Adaboost+Cascade Classifier.Experimental results on Multi-PIE and FDDB datasets show the good performance of the proposed method.2.A rapid face pose estimation algorithm is proposed to extract local features and classify face poses with a Random Forest algorithm.Two face pose estimation models are designed including coarse model and fine model.Five face poses are used in our coarse model:0 degree(frontal face),+45 degrees(half frontal face),-45 degrees(half frontal face),+90 degrees(profile face),and-90 degrees(profile face)along horizontal direction(yaw).Thirteen face poses are used in our fine model,each 15 degrees is a face pose.Our method shows good performance on Multi-PIE dataset.3.This paper proposed a Multi-ESR method based on Explicit Shape Regression method,in which different ESR models are trained in different views.The face pose estimation result is used in Multi-ESR to choose the corresponding model in different face poses.Experimental results show the proposed method yielding better performance comparing to the state-of-art methods handling multi-view face images.Furthermore,we offer a selection of Multi-ESR and trade-off Multi-ESR methods according to the different platforms.The proposed trade-off Multi-ESR method can reduce computational load without seriously affecting accuracy performance.4.A real system including face detection and an access control system based on face recognition is finished,which can automatically detect faces in images and videos,and recognize the authorized person through face recognition.