Facial Landmark Location And Its Application

With the development of computer vision and artificial intelligence,face recognition technology has been paid full attention by the researchers as an important branch of computer vision,and has been widely used in security,entertainment services,smart-home system and other fields.The whole process of face recognition consists of three parts:face detection,face alignment and face recognition.As the intermediate step of face detection and face recognition,face alignment can help us achieve better performance through aligning faces of different poses.Face alignment is usually based on feature point localization.Therefore,most of researchers take face alignment and facial landmark location as equal.The application of facial landmark location is not only limited to face alignment,we can develop some applications based on facial landmark locations for it can give the coordinate position of facial features.In addition,facial pose is also an important attribute of face,and there are many connections between facial pose estimation and facial landmark location.In this paper,2 methods of facial pose estimation are proposed,including rough estimation and fine estimation.The effect of facial landmark location algorithm is affected by illumination,occlusion and other factors.What's more,imbalanced training data will lead to poor performance when we deal with face images with large poses.To solve this problem,this paper presents an improved idea from the aspects of feature extraction,regression model selection and framework design.In addition,this paper fulfills a network structure of 5 landmarks location algorithm based on convolutional neural network,and verifies the effect of the network through experiments.Finally,the specific application based on facial landmark location algorithm is also given.Details are as follows:1.This paper proposes a rough pose estimation algorithm based on HOG feature and SVM classifier,as well as a fine pose estimation algorithm based on normalized coordinate value according to characteristics for facial pose estimation task.2.In view of the poor performance of LBF algorithm in dealing with images with large pose,an improved feature selection and algorithm flow design is proposed.In addition,the structure of a convolutional neural network is given,this network can locate 5 facial landmarks.Experimental results verifies the effect of the proposed network.3.The application based on facial pose estimation and facial landmark location algorithm is given by this paper.