Research On 3D Face Multi-attitude Synthesis And Application

Face recognition technology is one of the biometric recognition technologies involving multiple interdisciplinary subjects.To match and classify the face of a sample,the technology extracted the feature information from that face image.Due to the convenience and directness of the face image acquisition process,face recognition technology has been widely used in the fields of government,business,and public security.At present,many face recognition technologies have achieved good experimental results on ideal images,but in the process of application,the standard frontal face image acquisition is difficult,and the face image acquired under non-ideal conditions will have a big impact to the accuracy of recognition.In view of the problems brought about by environmental changes,especially face pose changes,based on the in-depth study of the current mainstream face pose correction method,this paper proposes a multi-pose correction algorithm based on 3D face model to synthesize face images in different poses into a front view,this algorithm can make the corrected output more robust,thereby improving the accuracy of face recognition.The main work and research results of this paper are as follows:1.This paper summarizes the research status and main challenges of multi-pose face synthesis.From the two aspects of two-dimensional images and three-dimensional models,the current research results and methods in the field of multi-pose face correction are elaborated in detail,then analyze and compare their respective performances,advantages and disadvantages.2.Based on the in-depth analysis of the current mainstream face pose normalization reconstruction algorithm,this paper proposes an algorithm for posture correction of multi-pose face images based on 3D model.The algorithm pre-processes the input image for the first,collects the feature information of the face.Then,by calculating a mapping relationship between the two-dimensional image and the standard 3D face model,to generate a three-dimensional feature point corresponding to the two-dimensional feature information,a 3D face model corresponding to theinput face is generated.Projecting the front view of the model onto the two-dimensional plane to obtain a preliminary positive face image,and finally to make the results more realistic,the invisible area on the preliminary result view is evaluated,and according to the corresponding visible features,the invisible area is filled by the mean-value coordinates(MVC)image editing method,so as to obtain the final face posture correction result.The experimental results show that the correction accuracy of the designed method can reach to more than 70%,and the face front synthesis in complex environment is better realized.3.Applied the multi-pose face synthesis method based on 3D modeling proposed in this paper to the research of gender judgment,age estimation and face recognition.The experimental results show that the multi-pose synthesis algorithm proposed in this paper has extensive practicality.Experimental analyzing show that the accuracy of the images generated by the synthesis of the algorithm in different contexts is significantly improved.