Design And Implementation Of Pirated Video Detection Scheme Based On Audio Fingerprint In Cloud Storage Environment

With the advent of cloud storage technology,multimedia information grows explosively.As more and more people prefer to watch videos online,a series of online video sites like Tudou,Youku,as well as some download softwares which provide "watch while downloading" function like Thunder,Baidu cloud,etc come out.However,a lot of loopholes have been found in the protection of video copyright.Although the country has put a lot of efforts into cracking piracy,detection of pirated videos technology is far more mature.Based on the background,we propose the pirate detection scheme based on audio word in cloud storage environment.In order to realize the scheme,we adopt the method of constructing audio dictionary based on multidimensional clustering,the method of audio sequence conversion based on feature extraction and the audio matching method based on inverted index.In this scheme,we first used the hyper-vector composed of many audio features.Secondly,we divide the continuous audio stream into short time audio segments with small fluctuation amplitude of characteristic value through multilayer audio segmentation method.Thirdly,we convert the short time audio segment sequence into an audio word sequence that can characterize the audio content by using the audio dictionary which is trained beforehand and establish inverted index.While retrieving,we convert the content that user submitted to audio word using inverted index to locate the candidate paragraph without traversing the database.Finally,we determine whether the video is pirated according to the relationship between the similarity of content and the threshold value of the candidate paragraphs.After the simulating experiment and the actual text,there is no obvious relation between three system performance index,average audio word sequence length,the retrieval time,the audio word computation time,and the video category,video lengths.The result is stabled in an acceptable range.On the other hand,the system can accurately retrieve audio and ripping audio.For noisy audio and audio with varied speed,better retrieval results can be obtained through longer sample time.The theme can greatly reduce the amount of storing data and computation time in audio retrieval.It also has good robustness to the attacking behavior.The final system retrieving accuracy can be stabilized at about 95%,which has the actual value of utilization.Hopefully,this scheme can provide an effective way for the copyright protection of other forms of electronic media data.