Geek question about p/b frames

[jwbrasil2]

Why isn't there a avisynth filter to debug only p or b frames? Since most digital noise do occur most on i frames pixels, why isn't there a reverse engineering filter able to apply only on those? I know some high-end software based on motion that can handle them. Like Deshaker, Mokey and Voodoo, they build planar points on each new geometric form coming on screen but they're not designed to remove noise. So it would be good to have a filter able to extract the background static picture first to a jpeg file, then add the i frames later so the background frames would be transformed into a matte painting and noise filters would only be applied to i frames (the main pixels).

[jagabo]

I don't think the encoded frame type is passed to AviSynth filters. And I think your premise is flawed anyway.

[poisondeathray]

Why isn't there a avisynth filter to debug only p or b frames? Since most digital noise do occur most on i frames pixels, why isn't there a reverse engineering filter able to apply only on those? I know some high-end software based on motion that can handle them. Like Deshaker, Mokey and Voodoo, they build planar points on each new geometric form coming on screen but they're not designed to remove noise. So it would be good to have a filter able to extract the background static picture first to a jpeg file, then add the i frames later so the background frames would be transformed into a matte painting and noise filters would only be applied to i frames (the main pixels).

Noise doesn't track very well. It's not predictable

Digital noise (or any noise) occurs on all frames across time. An encoded video stores the residuals in P and B frames. This means they store the differences between what is predicted. That's why noisy video is much larger, harder to compress and requires more bitrate. Denosing just I-frames doesn't automatically "fix" P and B frames, because in a noisy video - it's actually the noise and motion that constitues the most of the P and B frames residual. If noise were predictable, then the size of P/B frames would be very small (i.e. residual would be tiny)

When you decode the video (ie. view it), e.g use avisynth or any editing program - the concept of I,P,B or long GOP no longer exist. It's in the uncompressed domain, and you can think of everything as an "I" frame.

Long GOP is an encoding concept - it's a method of temporal compression. So for your approach to work, it would have to be applied at the time of encoding/recording, not after the fact.

[Gavino]

I don't think the encoded frame type is passed to AviSynth filters.

Agree it's not useful here, but note for future reference this information is available if using ffVideoSource.
See FF_PICT_TYPE under 'exported Avisynth variables' in the ffms2 documentation.

(Example use: http://forum.doom9.org/showthread.php?p=1546772#post1546772)