Design And Implementation Of Face Image Optimization Module Based On Weakly Constrained Indoor Video

At present,there are two kinds of face image acquisition method applied to face recognition system.One is the method which requires the person to be identified stands in front of the designated camera.This kind of method has the strong binding force to the person,and is only suitable for single face recognition,so its recognition efficiency is low.The other method is to obtain the face images directly from the surveillance video for recognition.Because the faces in the video are in an unconstrained state,there are lots of low-quality face images among the face images obtained from surveillance video.Deflection,blur and occlusion existing in face image lead to such low-quality face images.If these face images are used directly for recognition,the recognition accuracy will be reduced.In addition,redundancy between video frames greatly increases the burden on the system.In order to solve the problem that low-quality face images result in low recognition accuracy,this dissertation designed and implemented the face image optimization module.Before face recognition,image optimization will be done to face images obtained from surveillance video.It will not only improve the face recognition accuracy but also improve the operating efficiency of face recognition system.The main works of this dissertation are as follows:Firstly,a method combining face clustering and face tracking was proposed to generate the face image data.This method solved the problems that the processing procedure was complicated and the tracking object may be lost due to occlusion when face tracking method was adopted alone,and the problem that feature extraction was time-consuming when face clustering method was adopted alone.Secondly,according to the actual situation of the face state in the video,the face occlusion degree evaluation index and the face image clarity evaluation index were defined,and the method for determining the weight coefficients of the four evaluation index was proposed.A comprehensive evaluation index for evaluating the face image quality was obtained by linearly combing the four evaluation index.Finally,Python programming language was combined with OpenCV and PyQt to develop back-end application program and to design user interface.The face image optimization could be done through user interface and the optimized results could be displayed on the user interface.Compared with the method of face image data generation based on face tracking,the method proposed in this dissertation has improved about 15%in purity.Compared with the method of face image data generation based on face clustering,the method proposed in this dissertation has improved about 50%in time efficiency.Moreover,face recognition performance before and after face image optimization is compared through experiments.Experiment results show that the face recognition accuracy is improved by 13.2%and the total recognition speed is improved by 62s.