The Research On Algorithms Of Content-based Video Copy Detection

In today's information-based society,the network has become an indis-pensable part of people's lives.People share various forms of media(text,im-age,audio and video,etc.)through the Internet.Video is one of the network media forms.Some videos are created by video publishers,but some are re-leased without permission which are created by others,namely near-duplicate videos.So how to classify and manage the network videos,protect the copy-rights have aroused more and more attention,promoting the research in the field of Content-Based Video Copy Detection(CBVCD).Due to the continuous development of signal processing and transmission technology,more and more video transformations appeared,so there is a higher requirement for the robustness and discriminability of video features.In the past few years,although many video feature extraction algorithms have appeared,how to extract robust and differentiated video features under high efficiency is still one of the problems to be overcome in the research field.According to the above two problems,we put forward the corresponding solutions.Achievements are given as follows:It is found that the video feature extraction algorithms based on com-pressed domain is to extract the feature information of video after partial de-coding of the video stream.As one of the information of coded video stream,the motion vector is used to describe the motion information of the video mac-roblock.Although many researchers have tried to describe the content of the video through the motion vectors,but a lot of video feature extraction algo-rithms based on the motion vectors eventually transform the motion vector in-formation into the statistical histogram and lack the detailed description of the video structure information.In order to solve this problem,we put forward the concept of Motion Vector Imaging(MVI).Based on MVI,this paper proposes a Cascade system including MVI model and I frame model.Firstly,the video is partially decoded,the motion vectors of P frames and B frames are extracted,then the informations of motion vectors are mapped to the canvas to generate MVI.The I frames are fully decoded to pixel domain.The features of I frames and MVIs are extracted by Siamese Deep Neural Network(DNN).The exper-imental results show that the proposed video copy detection system based on compressed domain has good detection performance.Because a MVI is created by mapping the motion vectors of several re-lated frames,so it can only be used to describe the changing trend of video in a short period,it cannot be adopted to describe the intrinsic association of video frames over a long period.So,we propose another Sequence MVI(SMVI)model based on Long-term Recurrent Convolutional Network(LRCN)[1].This model regards SMVI as the input of LRCN.The video features extracted by LRCN can simultaneously describe the temporal and spatial information of video.Combined with the I-frame model proposed before,a Cascade system including SMVI model and I frame model is designed.The experiment proves that this system can detect the near-duplicate videos quickly and accurately.As for the problem of video feature matching,we propose a Multiscale Video Sequence Matching(MS-VSM)method,which can gradually detect and locate the similar segments between videos from coarse to fine scales.At the coarse scale,it makes use of the Maximum Weight Matching(MWM)algo-rithm to rapidly select several candidate reference videos from the database for a given query.Then for each candidate video,its most similar segment with respect to the given query is obtained at the middle scale by the Constrained Longest Ascending Matching Subsequence(CLAMS)algorithm,and then can be used to judge whether that candidate exists near-duplicate or not.If so,the precise locations of the near-duplicate segments in both query and reference videos are determined at the fine scale by using bi-directional scanning to check the matching similarity at the segments' boundaries.The MS-VSM model can accurately and rapidly locate the similar segments between video clips.