Double Compression Detection Of HEVC Videos With Different Bitrates

With the popularity of internet applications and the rapid development of the multimedia industry,digital video has gradually occupied a large part of our life,and it is also a critical component of judicial evidence.However,with the emergence of a large number of powerful video editing software,people can optionally tamper with videos.The spread of some incomprehensible malicious tampered videos has seriously affected social stability and hindered judicial forensics.Therefore,digital video forensics technology came into being.Because the tampered video must have undergone double compression,video recompression detection is one of the important branches of digital video forensics.HEVC(High Efficient Video Coding)is a new generation of video coding standard proposed in 2013.Compared to the old coding standard H.264,it is more efficient and it can support high resolution videos.In this paper,through going into core technologies of HEVC,from the perspective of HEVC prediction process,the characteristics of single-compressed video and double-compressed video under different bitrates are compared and analyzed.Based on it,two algorithms of HEVC video recompression detection with superior performance are proposed in this paper.(1)A HEVC video recompression forensics algorithm based on the prediction unit(PU)partitioning modes with different bitrates is proposed.Firstly,by analyzing the quadtree partitioning algorithm of HEVC standard,the PU partitioning modes of the first P-picture in all Group of Picture(GOP)are extracted,and then the histogram of PU partitioning modes is counted as the classification feature(the Histogram of PU Partitioning,HPP).Finally,the classification feature HPP is sent to the Support Vector Machine(SVM)for training and testing,thereby realizing the recompression detection of HEVC videos.The experimental results show that the proposed algorithm not only has high classification accuracy,but also has strong robustness against frarne-deletion,copy-paste and shifted GOP structure attacks.(2)A HEVC video recompression forensics algorithm based on prediction process with different birates is proposed.By comprehensively analyzing the prediction process of HEVC,it is found that the recompression has a great influence on the five prediction variables,namely:Coding Unit(CU)patitioning modes of I-picture,PU patitioning modes of I-picture,intra prediction modes of I-picture,CU patitioning modes of P-picture,and PU patitioning modes of P-picture.First,we extract these five variables in the video decoding stream,and then statistically calculate their probability matrix as single prediction features,and adopt three angles of feature fosion methods:feature fusion of I-picture series features,feature fusion of P-picture series features and full fusion of all prediction features.Finally,five single prediction features and three fosion features are sent to the SVM to complete the classification,respectively.The experimental results show that these five single prediction features are effective,they can distinguish HEVC single and double compressed videos,and feature fusion method can enhance the expressiveness of features and flirther improve the classification accuracy.In addition,the proposed algorithm is tested against frame-deletion,copy-paste and shifted GOP structure attacks.Experimental results show that the proposed algorithm is really robust.By comparing with previous excellent video recompression detection algorithms,it is verified that the proposed algorithm has great advantage in both the classification accuracy of recompressed video detection and the anti-attack robustness.