Order of filters ??

[Letricheur]

I am capturing home NTSC DV and converting it to either SVCD. I always resize (simple resize), deinterlace (field deinterlace())and smooth (temporal smoother) using avisynth, then encode in CCE. Sometimes I adjust the colours.

I have seen some comments about the order in which filters should be applied and wonder if anyone has an opinion on which is the best (fastest or best for quality) order in which to apply them?

[trbarry]

Generally you should always deinterlace or IVTC first as the other operations can make this not work as well. Exceptions might include first cropping to reduce the size and increase processing speed.

For the most part you can even horizontal resize first for speed if you have to with probably not too much detrimental effect. But any vertical resizing or blending before deinterlace should really be avoided.

- Tom

[Letricheur]

Wow, quick reply, thank you. I forgot to mention the 'crop bottom' I also do to remove the VHS info at the bottom of the screen (when I use Dazzle Hollywood Bridge DV capture/conversion on VHS material).

So it seems it would be best to use this order:

Crop Bottom (8)
Field Deinterlace()
Simple Resize(480,480)
Temporal Smoother(2,1)

[kayman]

Crop Bottom (8)
Field Deinterlace()
Temporal Smoother(2,1)
Simple Resize(480,480)


makes more sense you smooth before resize it will apear better


kayman

[Guest]

You might like to try TomsMoComp() for your straight video deinterlacing. FieldDeinterlace(), though usually adequate, is really specialized as a postprocessor for Telecide(). TomsMoComp() does some gentle spatial smoothing that might be useful to you, given that you apparently need some temporal smoothing.

[Boulder]

@Donald and Tom: Sorry, a bit OT here but I was wondering that should I use GreedyHMA or TomsMoComp to do the deinterlacing? I do a lot of TV captures and have to do pure deinterlacing before encoding to MPEG-1/2. I've used GreedyHMA(0,0,0,0,1,0,0,0) so far but is there any pros in TomsMoComp compared to that?

[droolian01]

Hello Boulder.

AFAIK, if you are doing straight tv captures (not dv mediated or anything to do with dv) you should use Telecide. If you see any problem frames using telecide then you almost certainly have blended frames, for which there is no real solution. I can say this as (i think) you live in a PAL country, and these blended fields occur with NTSC (US) tv programs.

I made the mistake of using tomsmocomp as a deinterlacer on my tv caps when it first came out, and the guys put me straight on this. DV captures (actually should they be called FILE TRANSFERS really??) are the definition of interlaced video, so tomsmocomp is ideally suited to this stuff.

Hope this is useful

[droolian01]

p.s.

OPPS - sorry! i forgot to mention that the reason i suggested this above was because you wrote -

I do a lot of TV captures and have to do pure deinterlacing before encoding to MPEG-1/2.

You should be field matching (not pure deinterlacing) tv caps, thats why i suggested telecide.

[Boulder]

Yes, Finland's a PAL country as well:) It's never occurred to me to try Telecide - my SVCDs and VCDs deinterlaced by GreedyHMA look fine to me but I guess I'll have to read the Decomb filter documents thoroughly..

[droolian01]

Hello again.

I think you may have made the same mistake that i did in thinking tv caps were 'interlaced'. Here is my 'eureka' moment (thanks to hakko) -

http://forum.doom9.org/showthread.php?s=&threadid=20032 (http://)

The way i figure it now is this. DV camera scans live (moving) scene and takes 50 fields per second, combining these into 25 frames, hence each field in each frame is sampled 1/50th later in time. this is pure interlaced and thus needs deinterlacing.

When TV studios transmit films their source is a 24 frame per sec PROGRESSIVE film. What they usually do is speed the film up to 25 fps then split each progressive (source) frame into 2 fields - this is what is then transmited as 25 fps 'interlaced'. But this is not the same type of interlaced as DV camera output as each field in the frame is sampled from exactly the same time. Therefore progressive frame recovery is possible (so you need to field match -> telecide).

It does get more complicated with 30fps NTSC US tv show transmitted on PAL - the dreaded blended frame - and i'm still getting my head around that.

If this was obvious - sorry - but someone may find this useful and have their own eureka moment lik i did!

See ya.

[Boulder]

Not obvious - very interesting! Thanks for sharing:)