Conclusion about Scene Change Detection

Refer to error concealment, there are lost of research paper. Today I want to conclude some paper about scene change detection.

Why scene change detection is so important? Because as the scene change happens, the concealment method for normal video is not suitable any more. Normal method uses temporal or spatial way to restore picture, but in this case, we can’t use temporal information to conceal picture. We are urgent to find another strategy to deal with it. But first of all, we should use available information to decide whether this picture has scene cut detection or not.

To go through many video sequence, I find that there is some scene change detection as below:

abrupt scene change, gradual scene change, flashlight, spinning, caption.

For these different kinds of scene change detection, there is a common method to detect: use MB information.

When the sequence has I、B、P frame, from the MB ratio of the picture, we can judge which change it is.

For instance, I1(P1)B1B2I2(P2), to calculate the intra MB ratio in I2(P2) and F, B, FB mode MB ratio in B1B2, we can set different threshold, and if it satisfies which domain, and the scene change kind it belongs.

When the sequence don’t contain B frame, we can use intra MB ratio in P frame and pixel difference in I frame. (Some guys come up with a method to compare the P frame’s MB type forward and backward I frame, this method has its limit, just think that the movement of two p frame is quite small, and two P frames will have most in common, but scene change happens)