Blending instead of interpolating in decomb?

[BluDChyLD]

I always use gordianknot for DivX encoding, and have always used the default decomb settings if deinterlacing is required. I've always found the "ghost" effect caused by field blending a little annoying but just took it for granted - I've only recently looked into decomb's settings in detail, and found out you can turn it off with the blend=false parameter :)

Are there any disadvantages to turning blending off, ie will it impact the quality of the deinterlacing?
Thanks

[sh0dan]

You might experience "flickering" pixels along sharp, almost static lines. In most clips it is not noticable however.
Another thing is that your video may not look as fluid, because the motion is not blurred as much anymore. But which one you prefer is more a matter of taste.

[wotef]

for PAL, you may even be able to get away with post=false

[mf]

Interpolating will show how bad Decomb is at detecting combs. Half of your picture will be interpolated for god knows what reason. I've tried loosening the comb settings but then I still have incorrect interpolation with only difference that some combs are let through.

[neuron2]

@mf

If you think you can make a better comb detector you should do it.

I see from other threads you are going around arrogantly trashing people's work. I find your tone offensive.

Something more objective people are aware of is that Decomb is designed for recovering progressive frames and the postprocessing is there only to fix occasional strays that get through the field matching. Therefore, the postprocessing is not designed to do as good a job as a general purpose video deinterlacer. If it was, the filter would become extremely slow. Sure, I could add motion map denoising, edge avoidance and other tricks, but do you really want a filter that runs at 2 fps?

Yes, FieldDeinterlace is also made available as a stand-alone filter, and many people find it adequate. But I've always acknowledged that for general video deinterlacing things like my Smart Deinterlacer and its Avisynth port can do a better job. But it comes at the expense of processing time. It's that old quality versus time tradeoff. There are lots of tools out there to let you select your point on the spectrum.

Did you know that many people are satisfied with solutions that effectively just discard a field? That is equivalent to interpolating the ENTIRE frame. So any adaptive approach that spares any static parts of the frame from interpolation represents a worthwhile gain.

[mf]

Originally posted by neuron2
I see from other threads you are going around arrogantly trashing people's work. I find your tone offensive.

1) Which threads ?
2) It wasn't meant like that. My apologies for that.
3) I never said I was nice :D.
Yes, FieldDeinterlace is also made available as a stand-alone filter, and many people find it adequate. But I've always acknowledged that for general video deinterlacing things like my Smart Deinterlacer and its Avisynth port can do a better job. But it comes at the expense of processing time. It's that old quality versus time tradeoff. There are lots of tools out there to let you select your point on the spectrum.
I've been "frustrated" with deinterlacers before I moved to AVISynth, and to be honest I actually presumed all your deinterlacers shared the same code to certain extents (would make sense, coming from one person). So again apologies for this wrong assumption, I just thought FieldDeinterlace was like an AVS equivalent of Smart Deinterlacer, although thinking back to vdub times I don't remember such bad comb detection. Anyway, unless a better MoComp comes along or someone wants to do a Hybrid Bob again, your filters are still the best :).

[neuron2]

Originally posted by mf
1) Which threads ?E.g., "Sigh- yet another bugreport: Crop and StackVertical in YV12".

I've been "frustrated" with deinterlacers before I moved to AVISynth, and to be honest I actually presumed all your deinterlacers shared the same code to certain extents (would make sense, coming from one person). So again apologies for this wrong assumption, I just thought FieldDeinterlace was like an AVS equivalent of Smart Deinterlacer, although thinking back to vdub times I don't remember such bad comb detection. Anyway, unless a better MoComp comes along or someone wants to do a Hybrid Bob again, your filters are still the best :). I think the mistake people make is thinking that perfect or even very high quality deinterlacing is even possible! Believe me, I've spent years researching it. There simply appears to be no really good way to reliably differentiate between combing and picture detail with the spatial frequency of combing. Interframe motion detection can be shown to be deficient and interfield motion detection is thwarted by the spatial offset of the fields.

I'm open to new ideas!

I am currently investigating a method whereby the impact of the spatial offset between fields can be ameliorated such that effective interfield motion detection can be leveraged. The danger as always is that it may be slow.

Regarding available deinterlacers, I find that TomsMoComp() and various Gunnar Thalin efforts are very good for general video deinterlacing. There is now an Avisynth port of Gunnar's area-based deinterlacer that you might find useful. What it adds to the FD approach is edge avoidance and (effectively) some motion map denoising. The edge avoidance, as indeed the base concept of interfield deinterlacing, are original ideas of Gunnar's. The motion map denoising is one of my ideas. MarcFD's aDeint() is also worthy of consideration.

[mf]

Originally posted by neuron2
E.g., "Sigh- yet another bugreport: Crop and StackVertical in YV12".

I think I was pretty nice and friendly in there. My starting tone wasn't all too good, but I corrected that and of course since the problem doesn't directly apply to me I could also just have not reported it :).

I am currently investigating a method whereby the impact of the spatial offset between fields can be ameliorated such that effective interfield motion detection can be leveraged. The danger as always is that it may be slow.

If that means "using spatial comb maps to control motion compensated deinterlacing" then I've been thinking about that as well :). It's just that I kind of gave up both on developing (finding motivated people when you're bad at coding yourself isn't easy ;)) as "proper deinterlacing" since every method has a catch. Which will only annoy me, which will only lead to me going ranting on IRC, which will only lead to bad forum posts in here and getting struck/suspended again. We don't want that to happen :p.
Regarding available deinterlacers, I find that TomsMoComp() and various Gunnar Thalin efforts are very good for general video deinterlacing. There is now an Avisynth port of Gunnar's area-based deinterlacer that you might find useful. What it adds to the FD approach is edge avoidance and (effectively) some motion map denoising. The edge avoidance, as indeed the base concept of interfield deinterlacing, are original idea's of Gunnar. The motion map denoising is one of my idea's. Yeah I've followed the Gunnar stuff :). It's nice to see all the choice in deinterlacers, compared to like uh, a year ago? :)

[Wilbert]

There is now an Avisynth port of Gunnar's area-based deinterlacer that you might find useful.
I guess I've missed this, do you have a link?

[neuron2]

Originally posted by Wilbert
I guess I've missed this, do you have a link? http://forum.doom9.org/showthread.php?s=&threadid=46161&highlight=areabased

[Druizk]

@neuron2

Congratulations...

I'am user, visited daily you Home site searching upgrades of your filters. Also reading this forum, and I think your filters are very very good utility,I have good results in my video-films. mf :mad: is ofensive don't kwnolege you work.

The last beta of Msharpen, Decomb and Dup are useful, neuron2 continue working, very user of Avisynth & VirtualDub(Mod) you have thanks.

By the way, in this moment is running your filter (Decomb y MSharpen) with 5 pass DivX 5.04 Beta, preview the second pass I'am surprise, is a old film DVD (NTSC - interlaced) with picture noise and blackness, but MSharpen and Decomb result is very good, new thanks....

[neuron2]

@Druizk

Thank you for your kind words.

We can overlook mf's abrasiveness. I've been known to suffer from it sometimes as well. :)

[mf]

Originally posted by Druizk
mf :mad: is ofensive don't kwnolege you work.
If that means "mf doesn't acknowledge your work", I just said his filters are good! And knowing that I don't kiss *** or be polite, that means I mean it.