Abstract:
A huge number of tasks can be performed by using forensic video processing such as frame
deletion recognition, frame addition recognition and identifying the origin of media file.
In our discussion, we are considering the case where some frames are deleted in the
video and we are going to detect it. As we delete a few number of frames the motion
vectors of the next frame increases in comparison to the previous frame. But the difficulty
in detecting this is that this case also occurs when there is fast change in events in the
video also.
In our discussion, we have proposed a technique to detect the frame deletion. For that
basically we have started in finding the motion vectors using block matching and then
finding the frame prediction error. The fingerprint which we got when the frames are
deleted is that as the frame from one GOP is predicted from another GOP. So, there is an
increase in residual error and it occurred at regular intervals. This effect can be observed
in the frequency domain.
Then we proposed an anti-forensic technique to hide these fingerprints by proposing a
frame prediction error pattern, Where there is no fingerprint in the frequency domain. To
get that fixed frame prediction error, we have the changed the motion vectors according
by which the video might not be changed much.
Finally, we have proposed a technique to detect this anti-forensic by finding the motion
vectors directly from the video by reading the bit-stream of video file and the other method
is by getting the motion vectors by frames using block matching algrithm. When we get
this parameter by both the methods there is difference in both of the motion vectors by
iv
which we can detect that there is frame deletion and anti forensics also applied.
The main contribution and learning of my work is the coding part which I did for
the above methods using OPENCV and FFMPEG in c++. A part from this the neural
network model which improved the efficiency.
