Face Alignment With Infrared And Depth Image

Face alignment is to detect facial landmarks from an input face image.Face alignment problem is an important problem in computer vision area,and many face-related applications,such as face recognition,face expression recognition and face tracking,depend on face align-ment results.In recent years,many face-related applications go from the laboratory to the market and come to our daily life.Face alignment algorithms play an important role behind those ap-plications.Therefore,from both a theoretical view and a practical view,face alignment is very important.Many face alignment algorithms are made for color images.Color images are easy to get,cheap to produce and familiar for researchers.Compared with color images,depth images provide completely different information.In depth images,each pixel represents the distance between the point in the scene and the position of the depth camera.Recently,with quick devel-opment of Kinect,many researchers start to look to depth images.However,few face alignment algorithms are designed for depth images,and even few face alignment algorithms are designed for the combination of depth images with other signals.We think that depth images can provide very different information from which provided by color images or infrared images,and the combination of them can give better result in face alignment.Hence,we decide to combine infrared images and depth images to do face alignment.In this paper,we design our algorithm based on Cascade Shape Regression model,and combine infrared images and depth images by feature extraction,feature selection and feature fusion.Experimental results show that this combination effectively exerts the best parts of these two kinds of signal,and gives better result than only using one kind of signal.We use this algorithm in our video-based driver fatigue detection system,and verify that it works correctly.