Improvement Of LBP Face Recognition Algorithm Based On Video Enrichment

With the rapid and widespread popularity of video surveillance,there is an urgent need for a user to quickly locate face information in a video surveillance file in a non-cooperating situation.Face recognition technology in concentrated video is undoubtedly the best solution.Concentrate the face in a large number of video surveillance,and then detect the face in the condensation,so that face recognition can be realized quickly from the surveillance video image..In practical applications,in terms of face recognition in video enrichment,under the influence of complexity and objective condition factors,there are many technical difficulties that need to be studied.The concentration efficiency and accuracy of video enrichment,as well as the efficiency and accuracy of face recognition,need to be solved urgently.In view of these problems,combined with their own reality,this paper mainly completed the following research work:1.In the condensed video,the key information of the moving target(the moving target studied in this paper is mainly the person in the video)is marked,so that the annotation information is displayed during the playing of the video,which is convenient for the user to view and locate.The moving target labeling method is to save the labeling information of all moving targets in a concentrated video frame and the playing time information corresponding to the frame as an annotation information packet;then,all the concentrated video frames are divided into media data packets,and the video frames are concentrated.The relative play time is the association between the media data packet of the frame and the tagged information packet,thereby implementing synchronization processing.Finally,the annotation information of the moving target of the concentrated video frame and the combination of the frame data are encapsulated and written into the concentrated video file,so that the correlation between the two is improved,and the annotation information can be conveniently saved and transmitted together with the video information.The annotation information of the movingtarget is directly displayed during the concentrated video playback.2.LBP(Local Binary Pattern)is multi-dimensional.Based on the multi-scale LBP extraction method of key points,this method can eliminate some unnecessary regional features.These useless features have no use for identification and may even affect the recognition effect.3.LBP submode algorithm.When LBP extracts such "features" directly from two pictures and performs discriminant analysis,it will cause a large error due to "position misalignment",and the introduced LBP sub-mode is improved.4.Based on the first two algorithms,a face recognition method based on multi-dimensional LBP sub-pattern is proposed.Combining the advantages of the above two,although the processing complexity and performance are improved;but in the case of face irregularity,the recognition rate is greatly improved;experimental data shows that the multi-dimensional LBP sub-pattern proposed in this paper has improved.Face recognition ability,and it is highly robust to facial expressions and positions as well as illumination changes.After market research and analysis,the moving target annotation method in the concentrated video and the improved LBP algorithm face recognition have practical application requirements in security.I have chosen a more in-depth research topic on the moving target annotation method,playback method and multi-dimensional improved LBP algorithm for concentrated video.