Chroma correction on VHS capture

[rds11]

I've been using the Camcorder Color Denoise filter (CCD) on most of my VHS captures. I found no better filter than this one to remove the chroma smear/trail that occurs on hockey and football games (on fast moving players).

However I came up with a situation where the filter is coloring something that should remain grey. I've included 2 videos, one without applying CCD and the other one with the filter applied. It's an hockey game, and the arena lighting is producing a shadow on the ice just below the board. Because the bottom of the boards are yellow, CCD is also coloring the shadow yellow, even when using the filter lightly (the pictures show where the issue is).

I've been using CCD on Virtualdub. I have not found a Avisynth filter that could remove the chroma smear/trail created by the moving players as good as CCD does it. So my question is, would it be possible to create a mask in order to apply CCD only on the non-yellowlish part of the chroma ?

[Sharc]

Sorry, but I can't see a significant difference between your 2 files, nor do I find your .png frames in the clips.
You could however apply MCDegrainSharp() on the fields which reduces both the color noise and the color smearing.

In avisynth something like

AVISource("Without CCD.avi")
converttoYUY2(interlaced=true)  #or converttoYV12(interlaced=true)
assumeTFF()
e=separatefields().selecteven().MCDegrainSharp()
o=separatefields().selectodd().MCDegrainSharp()
interleave(e,o).weave()

There might be better proposals though.

[jagabo]

You're running CCD on interlaced video? I don't think it supports that. And why are your videos RGB, not YUV?

[rds11]

You're running CCD on interlaced video? I don't think it supports that. And why are your videos RGB, not YUV?

I didn't know CCD does not support interlaced video. I have not noticed any artifacts from using CCD on interlaced material so far. I was under the impression that it was designed for interlaced material, given that the filter was created to correct chroma on old camcorders recordings (which I'm assuming were interlaced). I'm not using many Virtualdub filters other than CCD and Neatvideo, but is it possible to properly deinterlace on Virtualdub (like using separatefields() on Avisynth) before applying a filter and then re-interlace ? If not, how can I apply CCD correctly using Virtualdub ?

As for RGB, I always convert to RGB at the end of my Avisynth script. It's probably not required anymore, but I remember reading that some Virtualdub filters had issues with YUV videos, so I convert to RGB before doing anything in Virtualdub.

[rds11]

Sorry, but I can't see a significant difference between your 2 files, nor do I find your .png frames in the clips.
You could however apply MCDegrainSharp() on the fields which reduces both the color noise and the color smearing.

In avisynth something like

AVISource("Without CCD.avi")
converttoYUY2(interlaced=true)  #or converttoYV12(interlaced=true)
assumeTFF()
e=separatefields().selecteven().MCDegrainSharp()
o=separatefields().selectodd().MCDegrainSharp()
interleave(e,o).weave()

There might be better proposals though.

I've included 2 JPEGs to better illustrate the good/bad that CCD does. It's from frame 117.

[rds11]

I think I found a solution...

I imported CCD into Avisynth. I tried deinterlacing before applying the filter, but it created other issues with the shadow. Then, I found out a new version of CCD had been released a few months ago (I was using version 1.6, now it's 1.9).

I got he newest version from the developer's website :
http://acobw.narod.ru/file/ccd.zip

LoadVirtualDubPlugin("C:\Program Files\VirtualDub\plugins\ccd_32bit.vdf", "CCD", 0)
AVISource("Without CCD.avi")
assumeTFF()
e=separatefields().selecteven().CCD(30,0)
o=separatefields().selectodd().CCD(30,0)
interleave(e,o).weave()

I have not applied it to the whole video, but it works great on the clip I had attached. The smearing is removed at the same level as before, and the shadow is not being colored.

Also, I read that CCD was designed to work with RGB32.

Thank you guys for guiding me in the right direction.