Research Of 3D Facial Landmarks Automatic Location Based On Range Image

In the current research of 3D face recognition,automatic facial landmark detection algorithms are poorly developed.Manual calibration feature points is a heavy workload with using many time,and is cumbersome.So,a new automatic facial landmark detection algorithm blending 2D local corner points and 3D local shape index is proposed in this thesis.Firstly,a new method of face segmentation based on range image has been put forward,use integral projection and the anthropometric cranio-facial proportions,mark off eight landmark search windows about the canthus,ala nasi and angulus oris from range image,and then locate landmarks at evey windows.In the process of landmarks location,we first use Harris corner detector,extracts corner points with different parameters for different quality images,use corner points as an initial estimate of the location of landmarks.Then,do surface fitting at the neighbourhood of corner points,calculate the shape index,and get the final estimate of the location of landmarks.In order to text the validity of the method,we experiment on the Texas 3D face database,and used the standard deviation and average deviation with the true value landmarks to measure the location error.Among the 8 landmarks,the outer canthus have the maximum error with 1.958 mm about standard deviation,and 2.237 mm about average deviation.In addition to this,the confirmatory experiment about 3D face recognition base on the 8 landmarks has been carried out.The geodesic distances and euclidean distances between landmarks are used as geometric features to recognize faces.The recognition rate with rank=1 is 97.5% with neutral images,and 95.7% with expressive images.