Research On Detection Of Trans-coded HEVC Videos Based On PU Partition Modes

Owing to the development of Internet and multimedia technology,digital videos have been integrated into the public life and served as evidences in the areas of judicature,politics,news,business,and so on.However,open network brings in security issues which directly threaten the authenticity and integrity of digital videos.To solve the problem,passive digital video forensics emerges and is becoming one of the most important technical methods to identify the authentication of digital videos.Recompression is an inevitable process when tampering a digital video,which makes video double compression detection a popular research topic of passive video forensics.To the best of our knowledge,the majority detection methods are proposed to detect double compression under the situation that the same encoder is used during the consecutive compressions.However,few methods have been reported to detect double compression which adopts different encoders in the first and second compression.Thus a novel method is proposed to detect recompression videos that are trans-coded from Advanced Video Coding(AVC/H.264)encoder to High Efficiency Video Coding(HEVC)encoder.The main contributions of this these include the following three aspects.(1)The feasibility of applying Prediction Unit(PU)partition modes of P frames to detect trans-coded HEVC videos is analyzed.According to the phenomenon that blocking artifact caused by H.264 compression would lead to the change of PU partition modes of P frames when recompressing the videos with HEVC encoder,PU partition modes of P frames are adopted as distinguishing features to detect trans-coded videos.(2)A novel trans-coded HEVC video detection method is proposed based on PU partition modes of P frames.Firstly,the numbers of each PU partition mode in the first P frame of each Group of Picture(GOP)is counted.Then a 25-dimensional distinguishing feature set can be got by histogram.Finally the feature sets extracted from singly-compressed HEVC videos and trans-coded HEVC videos are sent into Support Vector Machine(SVM)for classification.The experimental results show that the method can effectively distinguish trans-coded HEVC videos from singly-compressed videos with the classification accuracy exceeding 90%in Quarter Common Intermediate Format(QCIF)and Common Intermediate Format(CIF)video datasets.(3)The robustness against frame deletion of the proposed method is further studied.The viability of applying the proposed feature set to frame-deleted trans-coding videos is explored by establishing a mathematical model.High classification results verify the strong robustness against frame deletion of the proposed method.