Research On Face Pose Correction Algorithm

Face recognition technology is the technology that uses computer system to analyze the information of the input face and extract its effective feature information to recognize the identity of the inputted person.Because of the non-contact characteristics and easy to collect and input,face recognition technology has been widely used in scenarios such as mobile phone unlocking and airport authentication.In the field of scientific research,In the field of scientific research,there's also a tremendous research fever.At present,a lot of fruitful work has been put forward,and excellent results have been achieved.Face recognition technology is often affected by such factors as posture,expression and illumination.However,many existing face recognition algorithms take the front face as input information.so how to correct the face pose to the front is a valuable research work.Based on the traditional face pose correction algorithm,in order to solve the problems in the research of traditional face pose correction algorithms: the distortion of local face area caused by rotation and loss of pixels,two improved face pose correction algorithms are proposed in this thesis.The main work is as follows:(1)Introduced the related concepts and commonly used technologies of face recognition,the definition of multi-pose face recognition and related research status,and the related face database used in the experiment.(2)Proposed a face pose correction method based on 3D Morphable Models(3DMM)to overcome the semantic inconsistency caused by rotation,i.e.local area distortion.This method can generate natural face images.Specifically,an adaptive3 DMM fitting algorithm is proposed to fit the input two-dimensional face image with the 3DMM.A Delaunay triangulation algorithm is proposed to deal with the local texture distortion in the fitting process.The face is divided into several triangular patches by using Delaunay triangulation,and then the triangular patches with pixel distortion are adjusted.Multiple experiments on LFW data sets show that the proposed method significantly improves the performance of face recognition.(3)Proposed a face pose correction algorithm based on image restoration to solve the problem of pixel missing after synthesizing face due to large deflection of face pose.Firstly,use landmark detection algorithm to detect landmark of non-frontal face image,and then use the attitude estimation adaptive 3DMM fitting algorithm to fit 2Dface image to 3DMM.Then,normalize the 3DMM to get the preliminary frontal face image,and then sent the frontal face image to the image restoration network based on depth neural network to repair the holes in the frontal face image which are missing pixels.Experiments are carried out from both quantitative and qualitative aspects.The experimental results show that the proposed algorithm can improve the performance of the classical face recognition algorithm compared with the traditional algorithm.In addition,it also shows that image restoration algorithm has broad research prospects in the field of face posture correction.