Research On Face Recognition Technology And Apply In Intelligent Video Retrieval

With the advance of "safe city",many public areas in the city have installed a camera to provide a powerful tool to combat crime,but how to quickly retrieve the target from the massive video data has become a problem,Video face retrieval technology came into being in this background,and the realization of its key technology is face recognition,how to improve the efficiency of face recognition and extract more accurate face expression information has become a hot and difficult research.This paper mainly studies and analyzes the face detection and face feature extraction process in face recognition.we train the classifier with CAS-PEAL-R1 face database which vary in pose,lighting,accessories and expression in order to solve the complexity of face detection in surveillance video,and then apply the classifier to video face detection system.First of all,single frame from video sequence is wiped off noise by the median filtering and average filtering,after that,the skin color segmentation of the preprocessed images was performed using the simple skin color model established in YCbCr space.We use geometric rules to exclude a part of face-like region in order to further accelerate the speed of face detection,and then use the classifier for the remaining face detection,and did some experiments in the static image and dynamic video.In the feature extraction stage,the original LBP operator,the uniform mode LBP,the rotation invariant LBP,the Haar type LBP algorithm and the PCA algorithm are analyzed and compared deeply,the traditional LBP operator can not extract the multi-scale and deep Hierarchical texture information and other issues,this paper presents an improved Haar LBP and PCA combined feature extraction method.Firstly,the facial images are expressed at two scales by wavelet decomposition,then the Haar LBP algorithm is used to extract the image texture features at each scale,the PCA algorithm is used to reduce the eigenvector,finally,the nearest neighbor classifier is selected to complete the classification and recognition of the face,and experiment on CAS-PEAL-R1,ORL and YALE database respectively.The experimental results show that the method in this paper has high robustness to human face change and speed up the face detection speed,the feature extraction method can extract more effective face information and improve the face recognition rate.System implementation phase,in the OpenCV and Qt development environment using C++ language programming to achieve a set of intelligent video face retrieval prototype system,selected three different levels of complexity of the video for the face search experiment.The experimental results show that the system can accomplish the real-time retrieval of the face,and the retrieval speed and accuracy can meet the practical application requirements,which can greatly improve the working efficiency of the personnel and reduce the work intensity.