HEVC Double Compression Detection Of GOP Non-aligned Videos

With the emergence of city video monitoring system,video filming equipment and editing software have earned its popularity.On the one hand,people will tamper with video content and then spread these falsified videos with the support of hardware and software,which will affect social stability;On the other hand,the digital video can already be used as the electronic evidence.If it is not guaranteed to be original,it will seriously jeopardize the judicial order.Therefore,video identification and falsification have gained attention from the society.Therefore,the research on video double compression detection is of great significance to video forensics.This paper mainly researches on double compression detection of GOP non-aligned videos.At present,research on the latest codec HEVC has not been widely concerned.Due to more flexible block encoding strategy in frames of HEVC,detection algorithms previously applicable to H.264 and MPEG are no longer efficiente.Hence,there are still many problems to be solved in this field.The innovations in this paper can be seen as follows:1)In this paper,an HEVC video double compression detection scheme based on SN-PUPM(Sequence of Number of Prediction Unit of its Prediction Mode)is firstly proposed.This algorithm mainly focuses on the abnormal periodic peak of the number of prediction units in different prediction modes in I-P frames.Based on this feature,Supportive Vector Machine is applied to detect the location of I-P frames and to distinguish the unique I-P frames from normal video frame.After that,periodic features of I-P frames are used in the process of periodical estimation and the parameter of GOP implied in the first compression process is obtained.Experiments show that the proposed scheme has better detection efficiency and robustness with different video content and video compression bitrate,especially for videos with relatively high bitrate and stable content,which is superior to other H.264 algorithms.2)This paper also proposed an HEVC video double compression detection scheme based on quantitative features of CTU structure division and prediction residual coefficient as PRD-BD(Prediction Residual Distribution with Block Distribution)feature.The scheme performs double compression detection for abnormal periodic spikes that may occur in the prediction mode,CTU structure and the distribution of predicted residual coefficients in I-P frames.Jenson-Shannon divergence is applied to construct the BD feature from prediction mode and CTU sturcture.Meanwhile,the extracted prediction residual is converted into PRD feature,thus a three-dimensional feature of PRD-BD is formed from the combination.Periodic estimation is adopted to analyze the periodicity of final feature sequence.Experiments demonstrate that this algorithm has higher I-P frame position recognition accuracies and show good stability.It is better than previous scheme based on SN-PUPM feature.The proposed detection schemes in this paper have good performance and practicability,and can be applied to digital video forensics,medical digital video identification and other fields.