Video Face Recognition Technology Based On PCA+SVM

With the development of times and science,video surveillance technology develops rapidly in modern society.Video surveillance is indispensable to the daily life of residents,the public security,and the detection of criminal cases in the public security system.As the core of video surveillance,face recognition has been paid more and more attention.As a new biometric recognition technology,face recognition has the advantages of easy collection,good visualization and high user acceptance compared with iris recognition,fingerprint recognition and other technologies.But face recognition still has the disadvantage of low accuracy and time-consuming.Due to the influence of various factors,the portrait recognition in the surveillance video is very different from the static portrait recognition.To a large extent,the face features in the video surveillance cannot represent the original information.The accuracy of face recognition in video monitoring directly affects the safety factor of users.Face recognition is faced with great challenges.For example,there are certain requirements on the place to collect information and we need to make a further exploration about the facial changes of people,the variety of people's expressions,etc.This paper mainly analyzes the algorithm based on face recognition in video surveillance.Firstly,the detection method of skin color and image likelihood degree for the face in video surveillance is detected;Secondly it proposes a more effective tracking algorithm based on MEANSHIFT algorithm,and finally puts forward a more effective method of face recognition by studying the block PCA algorithm+SVM algorithm based on principal component analysis.The main contents of this paper are as follows:1.Face image acquisition and preprocessing.Face detection in video surveillance is the first step and the cornerstone of the whole research.The skin color feature and detection method of image likelihood are analyzed,adopting the model based on YCbCr to detect.Some video backgrounds are close to skin color,which is not easy to recognize.So further binary operation is carried out on the image,and face aspect ratio is adopted to improve the robustness of recognition in this case.2.Face image tracking.In this paper,MeanShift is proposed,which can be updated adaptively based on the change of target features.Moreover,due to the interference of pigment-like regions,LBP texture feature combined with color histogram is proposed to improve the MeanShift algorithm.3.Face image recognition.The traditional method of face recognition is based on PCA and achieved by distance function,and the accuracy is relatively low.Therefore,on this basis,we proposed a PCA algorithm based on block,and added a SVM classifier,namely support vector machine.Compared with the traditional method,the recognition accuracy is greatly improved,from about 70%to about 90%,that is,the improved method is more convenient to use than the traditional method.