Using C3D with interlaced material

[numlock]

I have just captured some video tapes (regualar movies) and I'd like to use C3D to remove the noise. The original tapes were brand new so the picture is crystal clear. I will be encoding it to SVCD format and I want to keep it interlaced, because I want to view it ona TV set.

My first question is, should I expect any improvement in the quality since as I already mentioned the original material was excellent. Or should I just not bother with C3D at all ?

I assume I should use Convolution3d (preset="movieHQ").

My other question is, do I have to use

SeparateFields
Convolution3d (preset="movieHQ")
Weave

or just

Convolution3d (preset="movieHQ")

TIA

[stickboy]

If you don't use SeparateFields, the spatial component to Convolution 3D will smooth scanlines from different fields together.

If you do use SeparateFields, the temporal component will smooth scanlines from different fields together.

Tooting my own horn, I recommend using UnfoldFieldsVertical/FoldFieldsVertical (http://www.avisynth.org/~stickboy/) when using spatial-temporal smoothers on interlaced material:
UnfoldFieldsVertical(true)
Convolution3D(preset="movieHQ")
FoldFieldsVertical(true)

[numlock]

Thanks, I appreciate it.
One more question.

Shold I resize it to 480x480 before or after I apply C3D ?

Also, what resize function should I use there are like 4 of them ?
My source is Huffyuv 720x480

TIA

[paddycook]

What does the Setparity part of the UnfoldFieldsVertical function do, and when should you use it?

[stickboy]

SetParity sets the field-order of a clip; it's a complement to the internal GetParity function.

You shouldn't need to use SetParity yourself. It's just a function that UnfoldFieldsVertical/FoldFieldsVertical use.

numLock:
Resize afterward. I personally like LanczosResize, but BilinearResize will give you a more compressible (but blurrier) video.

[numlock]

stickyboy:

Somebody here me mentioned that by putting the odd and even fields apart from one another in the same frame, any information from motion which is shared between fields (which is very common) will be lost.

http://forum.doom9.org/showthread.php?s=&threadid=60369

Is this going to be a problem in my case when using C3D ?

[stickboy]

Yup, that's true; motion between fields is lost. Unless a filter is written to handle field-based material (like Lindsey's filters; I'm not sure about Convolution3D), I don't think there's really anything you can do about it without smoothing even/odd scanlines together. (Depending on the strength of the temporal filtering, though, using SeparateFields() ... Weave() might be good enough.)

[numlock]

So are there any other noise removing filters you can recommend that can handle field based material ?

[bairradino]

It seems that Convolution3D for Avisynth 2.5 doesn't work with YUY2 colorspaces.
Am I right?

[stickboy]

There's a YUY2 version for AviSynth 2.5 available from WarpEnterprises's filter page (http://www.avisynth.org/~warpenterprises/).

[JuanC]

Originally posted by numlock
So are there any other noise removing filters you can recommend that can handle field based material ? PeachSmoother is great and very fast! It is aware of the field/frame based flags and will process your clip accordingly. Look at this post (http://forum.doom9.org/showthread.php?s=&postid=365102#post365102) for recommendations from the author on how to use it. And this thread (http://forum.doom9.org/showthread.php?s=&threadid=58674) to download the avisynth 2.5 version of the filter.

BTW, it's more accurate to call it noise reduction. If you prefer, you can get the original version (with the help files) for avisynth 1.x/2.0.x here (http://www.avisynth.org/~warpenterprises/)

[vmesquita]

This is what I use to do Convolution3D on interlaced material:

a=AVISource("1.avi")
a=a.Converttoyuy2()

a.SeparateFields()
even=SelectEven(a).Convolution3d(0, 10, 17, 10, 13, 2.8, 0)
odd=SelectOdd(a).Convolution3d(0, 10, 17, 10, 13, 2.8, 0)
a=Interleave(even,odd).weave()
return(a)

I use it mainly for my satellite captures. PeachSmoother is good for analog material, like VHS.

[]'s
Vmesquita

[ADLANCAS]

@Vmesquita,

I used to do what you described, but now I use:
AviSource("D:\Teste\IUvcr\capture.avi")
dgbob(1)
##########
#Filter
##########
separatefields()
selectevery(4,0,3)
Weave()

It´s slower, but quality is much better.

Alexandre

[Piper]

I'm curious, was running C3D on interlaced material meant to improve quality before deinterlacing or are you not deinterlacing at all?

The reason I ask is, I'm also looking for an optimal way of using C3D on my sources, but I deinterlace my source almost 100% of the time.

I had been using
SeparateFields()
Convolution3D(preset="movieHQ")
Weave()

KernelDeint (etc)

After reading this thread, I'm reconsidering my script.

[bb]

Piper,

if you deinterlace your video, there's no need to separate the fields. Just deinterlace first, then apply Convolution3D.

bb