An idea to improve Decomb's telecide detection using VThresh always, frame, likely

andy

As I was looking at ways to improve my Decomb settings for catching anime combing issues in IVTC, I noticed a pattern with Vmetrics.

As I understand, Vmetrics tell us how much the frame is combed.  Thus, a higher Vmetric value means the frame is more likely combed.  In my anime material, the frame that should be chosen ALMOST always had the lowest Vmetric value.  This even allowed proper duplicate frame detection in my source.  Now, I noticed that telecide does not always select the frame (either current, next, or previous) based on the lowest Vmetric.

Is there a reason for this and why?  One big problem that is hard for me to compensate for in Decomb is telling it to duplicate a frame because the next one according to Vmetric is interlaced regardless of it being current or next.  Therefore, choosing previous in this case works.  Although I noticed that most of the times, choosing the one of the next set of frames works too.   

Now, this may seem obvious and already noted, but I didn't see how this could be implemented with the current Decomb.

Also, it is rare, but sometimes, both c,n, or b have the same Vmetric.  Then, the one with the highest Mmetric should be used for the clean progressive frame.

In all of these cases, sometimes selected the P or previous frame is best and should be checked.

Also, in decimation, the one with the lowest V or D metric can also aid proper decimation of near duplicates with interlaced artifacts.

If I am not making sense, I will try to elaborate/clarify accordingly.

-Dali

Isn't the p matching covered by bthresh=(slightly higher than vthresh?) and back=1, or 2?

Upon the sticky world of anime, an interlaced frame can have a higher or lower vthresh than a progressive frame which happen to be the following frame. One such example is the well-known interlaced mouthing.