Research On Multi-pose Face Recognition Based On Key-frame Of Video

Face recognition is a hot research field of pattern recognition,and it is also a popular technology of biometrics.At present,some related products have been used in commercial activities.In recent years,mobile smart devices are growing rapidly,especially mobile devices based on Android can be seen everywhere,the research of face recognition for mobile devices is also rising with the development of smart mobile devices and mobile terminal,the greatest difficulty of the mobile terminal recognition is that the processor computing power is relatively weak,and mobile devices generally require real-time recognition,so the PC-side algorithms are not suitable for mobile devices.The main work of this thesis is to propose a real-time face recognition scheme on the mobile platform,specific work are as follows:(1)According to the characteristics of mobile devices and the low recognition rate of face recognition algorithms currently applied to mobile devices,in this thesis,a multi-pose face recognition method based on key-frame of video is proposed.The video contains rich face information,it can get different light and poses of the face image,so it can effectively improve the recognition rate.The video can make full use of the user's multi-pose image and confirm the face recognition results,so as to ensure a better accuracy,and the combination of multi-pose can effectively prevent the use of photos and other deceptive behaviors.(2)This thesis analyzes the common methods of face detection,and uses the method based on Adaboost for face detection in video stream.In this thesis,the interval frame method is proposed for face detection and recognition.A suitable interval frame number is set,which makes sure to reduce the number of frames as much as possible while the recognition rate is in better circumstances.The face detection confirms the eye,nose,mouth sequentially to screen face images.It can speed up the detection speed through adaptive adjustment of the search scale and exclude unqualified face images effectively.The face tracking determines the location of the face through tracking the eyeballs and helps positioning the eyeballs of next frame.(3)On the basis of the research on the methods of feature points location and extraction in face recognition,this thesis proposes an authentication scheme that combining geometric features and local features.The scheme is validated on key frames obtained from face detection and screening.Firstly,advantages of the mobile device itself with the clear images and the user cooperation is used to locate and extract geometric features,and then detail features is matched through the improved template matching algorithm,which makes up the disadvantages of geometric features that losing detail features.(4)Experiments are carried out on the mobile device to verify the effectiveness of the proposed scheme and algorithm.The experiment results show that the scheme and algorithm improve recognition rate and real-time performance,and have practical application value.