Multi-granularity Hashing Based Method For Video Authentication And Tampering Localization

With the rapid development of multimedia technology,the high-definition digital camera and video editing software have been widely used in practical application,which make it more easy and convenient to process the video's content,including adding,deleting,modifying the objects in the video,and blurring edge of the object to make it uneasy to detect by human eyes.These video processing tools are easy for an attacker to obtain a video and edit it to mislead other's understanding,which reduces the credibility of videos we obtain.Video authentication and tampering localization are important in news media,court,and public security.Thus,an automate method is of great important and video hashing is one of the most popular methods.Video hashing extracts the features of the video as a hash vector,which is robust against tolerant operations,sensitive to tampering operations and is able to locate the tampering.In this paper,we propose a method for video authentication and tampering localization,which includes the following works.Firstly,we introduce the research background and significance of video authentication technology.Then we evaluate the advantages and disadvantages of state-of-the-arts in this field,and summarize the common problem of state-of-the-arts.That is,they are hard to balance the detection requirements: robustness,sensitiveness and forgery localizable.Then,to solve this problem,this paper proposes a new algorithm on the video authentication and forgery detection based on multi-granularity hashing algorithm.The contributions of this algorithm are fourfold:(1)extract video hashes from multiple granularities to balance the robustness and sensitiveness;(2)incorporate 3D-DWT and PCET to extract geometric invariant spatio-temporal features;(3)a geometric correction technique based on the hash of angular blocks is proposed to locate the tampered regions in the rotated frame;(4)a coarse-to-fine tampering localization strategy is proposed to improve the detection accuracy.Experimental results show that compared with state-of-the-art methods,our method is more robust against content-preserving operations.It is particularly robust against temporal de-synchronization and geometrical transformation,and has high tampering localization accuracy even when the video is rotated.Lastly,we summarize the main work of this paper and point out the future work of the proposed algorithm.