Aiming at the problem that in the process of video face replacement

face detection and feature points localization do not consider the contact of consecutive frames

in time domain which causes jitter and misplacement of the processed face in video

and the method of face registration in video based on alignment metric is put forward. The candidate region of the unregistrated face image and the facial features of the reference face image are determined by the feature point positioning

and the reference face search kernel is obtained according to facial gradient information. The image scale Pyramid is constructed

and search the extreme value of the alignment metric in the scaled image Pyramid at a high speed; the coordinates of the extreme value is the corresponding matching point that get a highest similarity compared with the search kernel and the matching pair of points are generated.Procrustes analysis algorithm is applied to calculate the transformation matrix of the corresponding matching points and the affine transformation of the image to be registered is completed. According to the experimental data

the speed of the optimization search is improved

and compared with other algorithms

the aligned face position and pose of the frame is much more stable in the processed face video by the proposed algorithm. Meanwhile the face location in the processed videos is naturally smooth when playing. The algorithm eliminates the influence of the skin color region

uses the valid face information of consecutive frames

maintains the position and poses of the face area with large gray scale change

and realizes the precise registration of the video face.