Video Retrieval Research Based On Contourlet Transform In Digital Libraries

With the rapid development of network technology and the popular of multimedia devices,as a result,digital video resource is becoming more and more important in digital libraries.As one way to transfer information which combined audio and pictures,video can maximize the display of the real situation and the incident scene.To match with text and sound,video is more real and more suitable for the charateristics of the information needs of people,video is favored by the public.Faced with the huge amount of video resources,how to make the digital libraries users quickly and accurately find the required video information has become an urgent problem to be solved.User demand can not be satisfied with the traditional method of keywords-based retrieval.In order to compensate the lack of traditional method,content-based retrieval method was proposed,which could retrieve the video content directly and also could improve the retrieval accuracy.Therefore,this dissertation focuses on video retrieval in digital library and presents a content based video retrieval framework for digital library.Combining with the Contourlet transform theory,this dissertation describes an approach to video retrieval,in order to provide a lot of valuable reference for digital library video resources research.The main research work in this dissertation are concentrated on three main parts: video resources construction in digital library,video retrieval framework and video retrieval scheme based on Contourlet transform.The following are the main methods used in this dissertion: Literature Survey Analysis,Bibliometric,Simulation,Comparative analysis and Multi-scale Transform.According to the literature survey and analysis,the dissertion concludes that: the way of video resources construction in digital library is self-building?purchasing and sharing.Purchasing is the main way and self-building is for special local resource.In the procedure of video resources construction,there are two trendency emerged that is popularization and outsourcing.This dissertion,selected 5422 relevant articles about video retrieval from the databases of Web of Science as its sources,aided by the software of Cite Space,makes a visualized analysis in temporal distribution,national power,core institution,core author,crucial author,intellectual base,research hotspot and research front.The conclusions of this study are that video retrieval research is still in growth period,the mian power is distributed in America and China,India has huge development potential,the important author is David G.Lowe,Josef Sivic,Cees G.M.Snoek and Chang Edward.The intellectual base of video retrieval research can be traced back to As We May Think,which was published in 1945 by Bush.The latest research hotspot is 3-D objective retrieval,big data and deep learning.The research front is video content analysis,the application of video retrieval and video retrieval system construction.Combining with typical digital library video retrieval framework,this dissertion constructs a video retrieval framework based on Contourlet transform.The presented framework is used C/S architecture and composed of three modules,including video management,video analysis and user interface.This dissertion focus on video analysis module which contains the procedure of key frame extraction,feature extraction,similarity match and so on.Aimed at key-frame extraction issue,this dissertion adopts fixed frame method,which take one frame every 10 frames as the key-frame.This dissertion adopts Contourlet transform to decompose key-frames into low-frequency subband and high-frequency subband.Then we take mean and variance attribution of high-frequency subband to form the feature vector for depicting key-frame.The video finger is composed by key-frame feature vector.The similarity match uses Euclidean distance to measure the similarity of videos.In the end of this chapter,the simulation experiment is carried out to verify the presented video retrieval framework.10 video files are selected from the VCDB database,and the selected videos are attacked in 10 ways,and the 100 video files are generated as the experimental data.The dissertion uses the accuracy,recall and F1 value as the evaluation criteria for the retrieval results.The algorithm is achieved in Matlab2013 a,and the video retrieval based on Wavelet is set as a comparative experiment.The experimental result show that the Contourlet based video retrieval scheme is superior to the Wavelet based video retrieval scheme based on robust to various attacks.The experimental result indicates that the video retrieval framework based on Contourlet transform for digital library is feasible and effective.The system can realizes the retrieval of video content,which is helpful for the digital library to improve the service level of the video retrieval and make up for the deficiency of the keyword-based retrieval.