Research On Noise-level Based Forensics For Tampering Of Video Frame Rate Up-Conversion

With the development of information technology,the digital video has become a widely used information dissemination carrier.However,the digital video is also facing the danger of being tampered with,and the authenticity,security and integrity of its information content are getting more and more attention.Frame Rate Up-Conversion(FRUC)is a common method for video forgery.In this thesis,the forensics method is studied for FRUC tampering.Especially,the method based on noise level estimation is proposed to detect frame repetition and Motion-Compensated FRUC(MC-FRUC)tampering.As the evidence of detection,the noise level is first estimated by Median Absolute Deviation(MAD)of wavelet coefficients,then its temporal periodicity is detected,and finally the video is automatically detected to determinate whether it is forged by a hard threshold decision.The following describes the main research achievements of this thesis.(1)A passive forensics algorithm for frame repetition tampering is proposed.Since the noise always exists in the video frame,the existing algorithm forensic performance is easily disturbed by noise.However,when the video is forged by frame repetition,its noise level will also change.We use Gaussian distribution to model the noise in video,and detect the frame repetition tampering by observing the temporal varying of noise level.The experimental results show that the proposed algorithm can overcome the noise robustness and can automatically detect frame repetition tampering.(2)A passive forensics algorithm for MC-FRUC tampering is proposed.Although frame repetition can achieve better visual effects for the video static region,the jittering or ghosting artifacts often occur because the motion between successive frames is neglected.In order to improve the video quality,more advanced MC-FRUC is adopted.At present,there are few forensic algorithms for MC-FRUC tampering.Therefore,it is necessary to forensic MC-FRUC tampering.When the video frame is interpolated by MC-FRUC,the noise accumulation will occur.Then,the noise level has certain feasibility as the detection evidence,and MC-FRUC tampering is detected by analyzing noise level changes in time domain.The experimental results show that the proposed algorithm has low complexity and good forensics performance.(3)An active forensics algorithm for MC-FRUC tampering is proposed.Since the original video includes the noise of the device and external noise,the actual Gaussian distribution modeling does not fully reflect the statistics of noise.This affects stability of passive forensics algorithms based on noise level estimation.The proposed algorithm uses Gaussian distribution to model the noise which is pre-mixed Gaussian noise with known statistics.We analyze the varying noise level in time,and use the hard threshold decision to automatically identify MC-FRUC tampering.The experimental results show that the proposed algorithm has good detection performance for different MC-FRUC methods.