Color bleeding on an anime VHS

[wubikens]

On the edges of the red animal fur, and on the edges of the blue jacket over the red animal fur, there's some color bleeding. Tried a few different things to get rid of it, such a lot of these Chroma Correction filters: http://avisynth.nl/index.php/External_filters#Chroma_correction

It doesn't necessarily seem to be a chroma issue though, as even though I can shift the chroma, the colors still bleed in the same spots without moving. I've tried this AI model: https://openmodeldb.info/models/1x-BleedOut-Compact

And this worked the best, as it actually corrected the color bleeding, but it negatively affects the video in other ways. So it isn't 100% reliable.

Would anyone have any suggestions? Thank you.

[jagabo]

I'm not at a place where I can view your sample but VHS always has chroma bleed problems. It only has about 40 lines of chroma resolution across the full width of the frame. The (partial) fix is to resize to a small width, sharpen horizontally, then upscale the width with a sharp upscaler. Then merge the new chroma with the original luma.

[wubikens]

I'm not at a place where I can view your sample but VHS always has chroma bleed problems. It only has about 40 lines of chroma resolution across the full width of the frame. The (partial) fix is to resize to a small width, sharpen horizontally, then upscale the width with a sharp upscaler. Then merge the new chroma with the original luma.

Got it! Thank you so much for the suggestion, I'll give that a try.

I'm not entirely sure which filters to use, but would you have a sample script in mind that I could experiment with by chance?

[Selur]

but it negatively affects the video in other ways.

Can you elaborate on that? (maybe share a sample where it causes the problem?)
To me, shifting the chroma by -4 (might be a bit much) and using BleedOut does seem fine on the sample.


Cu Selur

[Selur]

It only has about 40 lines of chroma resolution across the full width of the frame.

That sounds strange. A line should be a row, so shouldn't it be ~40 columns, not lines of chroma resolution?
iirc. (a bit fuzzy about that), it was something like:
NTSC VHS (effective):
Luma: ~240×480
Chroma: ~40×240
PAL VHS (effective):
Luma: ~240×576
Chroma: ~40×288
This was since VHS encodes chroma using a very low-frequency subcarrier (~629 kHz NTSC). The bandwidth for chroma is limited to about 500–800 kHz, compared to 3–4 MHz for luma. (side note: it's still correct to capture at a width of 720 to keep all the analog bandwidth.)

Cu Selur

[Sharc]

... The (partial) fix is to resize to a small width, sharpen horizontally, then upscale the width with a sharp upscaler. Then merge the new chroma with the original luma.

Got it! Thank you so much for the suggestion, I'll give that a try.

I'm not entirely sure which filters to use, but would you have a sample script in mind that I could experiment with by chance?

Try something like (on progressive video):

v=<source filter>
Spline36Resize(v,width(v)/4*2, height(v))  #downscale width mod2
aWarpSharp(depth=25).Sharpen(0.8,0.2) #sharpening
nnedi3_rpow2(2,cshift="Spline36Resize",fwidth=width(v),fheight=height(v)) #restore the original dimensions
mergechroma(v,last) #apply the sharpened chroma
chromashiftSP(-2,0) #for this particular kon.avi clip only

[jagabo]

It only has about 40 lines of chroma resolution across the full width of the frame.

That sounds strange. A line should be a row, so shouldn't it be ~40 columns, not lines of chroma resolution?

I meant "line" in the sense that you measure horizontal resolution with vertical "lines" (and vertical resolution with horizontal "lines"). So we're saying the same thing.

[Selur]

ah, okay.

[wubikens]

... The (partial) fix is to resize to a small width, sharpen horizontally, then upscale the width with a sharp upscaler. Then merge the new chroma with the original luma.

Got it! Thank you so much for the suggestion, I'll give that a try.

I'm not entirely sure which filters to use, but would you have a sample script in mind that I could experiment with by chance?

Try something like (on progressive video):

v=<source filter>
Spline36Resize(v,width(v)/4*2, height(v))  #downscale width mod2
aWarpSharp(depth=25).Sharpen(0.8,0.2) #sharpening
nnedi3_rpow2(2,cshift="Spline36Resize",fwidth=width(v),fheight=height(v)) #restore the original dimensions
mergechroma(v,last) #apply the sharpened chroma
chromashiftSP(-2,0) #for this particular kon.avi clip only

Thanks! I'll play around and give that a try.

but it negatively affects the video in other ways.

Can you elaborate on that? (maybe share a sample where it causes the problem?)
To me, shifting the chroma by -4 (might be a bit much) and using BleedOut does seem fine on the sample.


Cu Selur

Absolutely, I'll upload two samples of a different scene: one with Bleedout and one without.

In the one without Bleedout, you can see some red chroma/color bleeding around both characters. In the one with Bleedout, you can see the red chroma/color bleeding gets pushed onto other parts of the characters that aren't red (notice the faces and skin tone changes especially): https://imgsli.com/Mzc2MDAw

[Selur]

Have you tried applying a chroma shift before bleedout?
That does seem to lessen this effect: https://imgsli.com/Mzc2MDM2
(alternatively, apply a chroma shift to all frames and bleedout only to some scenes)

Cu Selur