Decomb 5.2.1 released

[Guest]

I've made an update to Decomb.

Decimate modes 0, 1, and 3 now use the same windowed comparisons as mode 2 does. This is the strategy used in Dup() and it gives more accurate and reliable decimation. Somehow, it got omitted from 5.1.1.
:eek:

Please get it from my web site.

[SoonUDie]

Hmmm, weird. Glad to hear it's fixed :)

Also, have you considered making a version of dup that works with the individual blocks, not just the entire frame? You could do something like what dup does now, except use neighboring blocks to determine whether to update a particular block or not. Might or might not be worthwhile due to block-like artifacts (but definately higher compressibility).

[Guest]

Originally posted by SoonUDie
Also, have you considered making a version of dup that works with the individual blocks, not just the entire frame? That was discussed back in the original Dup() thread. We just concluded it might be worth trying, but it was never a priority for me to implement it.

[APF_Gandalf]

a "simple" way to try if it worth implementing a "blockdup" is to crop the entire movie in 8~8 (maybe 16~16) parts, apply dup and then stack the parts back to their original place.

[SoonUDie]

Originally posted by APF_Gandalf
a "simple" way to try if it worth implementing a "blockdup" is to crop the entire movie in 8~8 (maybe 16~16) parts, apply dup and then stack the parts back to their original place. I actually did that as an experiment. It worked, but the gain was small, and it was hideously slow.

[sysKin]

Hi,

My knowladge about deinterlacing/ivtc is none, I didn't have to do it until today. Today, however, I finally decided to do something about one horribly telecined music video. I read (among other things) decomb's manual and I have a suggestion.

Decimate(mode=0) is what I'm using (after telecide), as my source originated from film. Now, the manual says that it detects two most similar frames and discards second of the two.

How about the ability to:
- discard the first of the two
- blend the two frames into one, which will halve the noise if we're lucky

From what I see in my clip, quite often one of these frames has some ghosting while the other doesn't. The choice between the 3 methods could help, or at least be useful, with removing the ghosting as much as possible.

If what I said is rubbish then excuse me for wasting your time ;)

thanks,
Radek

[Guest]

Originally posted by sysKin
How about the ability to:
- discard the first of the two
- blend the two frames into one, which will halve the noise if we're lucky

From what I see in my clip, quite often one of these frames has some ghosting while the other doesn't. The choice between the 3 methods could help, or at least be useful, with removing the ghosting as much as possible.
The ability to discard the first would be useful only if the blends are always in the first of the two. I can't see what process guarantees that that is the case.

Blending for noise reduction is not a proper decimation function, in my view. You can use the Dup() filter (written by myself and sh0dan) to achieve that functionality.