Face Alignment And Its Application Based On Multi-template SDM

Face alignment is the process of locating key points on the face image.Generally speaking,face alignment process can be divided into two steps:face detection and locating key points.Face alignment can get accurate locations of feature points and increase the reliability of feature extraction.Face alignment is an essential part of many face-related tasks and has been a hot topic in the field of face information processing.Applications such as head attitude estimation,face recognition and emotional analysis should be based on face alignment.In real life,face alignment has been widely used in public security,video surveillance,intelligent city,electronic payment,and so on.Face alignment technology has been developed rapidly in recent years,but it is still difficult to develop an efficient and stable face alignment system.Some of problems are fairly apparent,such as the stability and efficiency of face detection,and the ability to resist interference of light and shelter.In this thesis,we use Viola-Jones face detection algorithm to detect face for face alignment,which includes the following important parts:1)Using Haar features to describe face image;2)Constructing integral image and quickly geting several different rectangular features;3)Using Adaboost algorithm for training;4)Grouping features and establishing hierarchical classifiers.Supervised Descent Method(SDM)is a highly efficient and accurate approach for facial landmark locating and face alignment.In the training phase,it learns a sequence of descent directions to minimize the difference between the estimated shape and the ground truth in feature space.Then in the testing phase,it utilizes these descent directions to predict shape increment iteratively.However,when the face expression or direction changes too much,the general SDM cannot obtain good performance due to large variations between the initial shape and the target shape.In this thesis,we propose a multi-template SDM(MtSDM)which can maintain high accuracy on training data and meanwhile improve the accuracy on test data.Instead of constructing only one model in the training phase,several different models are constructed to deal with large variations on expressions or head poses.And in the test phase,the distances between some specific landmarks are calculated to select an optimal model to update the point location.The experimental results show that our proposed method improves the performance and performs better than several existing state-of-the-art methods.We apply the MtSDM to fatigue driving detection successfully and obtain promising results.