90's Japanese anime VHS clean up attempt

[obaku]

Hello everyone,

I recently captured a NTSC-J VHS using a JVC SR-V10U (TBC on, Norm, Digital R3=off) and a Happauge USB Live 2, (No external TBC sadly, it's hard to find).
Then i tried to clean it up a bit using Avisynth and Virtualdub, here's what i've done so far:

Samples: https://mega.nz/folder/hpBR2YyY#y9CYw3g25BmF4IW0NJHHrQ
[/URL]

DirectShowSource("G:\Record\sjg2.avi", pixel_type="YUY2")
ConvertToYV12(interlaced=true)
ChromaShift(l=0,c=4)
DeHalo_Alpha(rx=1.7, ry=1.5, darkstr=0.1, brightstr=1.0, lowsens=60, highsens=45, ss=1.0)
MCTemporalDenoise(settings="low")
LSFmod(defaults="slow", ss_x=1.00,ss_y=1.0,secure=true,strength=40)
ConvertToYUY2(interlaced=true)

[Attachment 53430 - Click to enlarge]

It's not much, i know there won't be any miracle so i don't want to "overfilter" it (and i kinda think i already did)

Is there something i should change ? I'm new to this so i thought asking experienced users for some tips would be a good idea.

Thanks in advance !

[LMotlow]

No one is going to spend half an hour or longer trying to navigate that download site for gigabytes of video. Make shorter samples, before processing not after.

You won't get far working with anime in Virtualub alone. You need Avisynth.

[KarMa]

No one is going to spend half an hour or longer trying to navigate that download site for gigabytes of video. Make shorter samples, before processing not after.

Looks pretty similar to and operates like GDrive. It's fine.

[lingyi]

I can't remember the name of the member, hopefully someone else will and post it, but there's a guy with a DragonballZ icon that works with a lot of anime. Once this name is posted, check out his threads and possibly PM him to get him on this thread.

[lingyi]

killerteengohan is probably the poster I'm thinking of. Search through his posts https://forum.videohelp.com/search.php?searchid=2094322 as he posts a lot about anime.

[KarMa]

JVC SR-V10U (TBC on, Norm, Digital R3=off)

I would strongly urge you to re-record your video in Edit mode. As Norm is just going to smear everything.

[jagabo]

You are using a bunch of progressive filters on interlaced video. That's sure to screw things up. Also, your video is TFF but AviSynth assumes it's BFF. In this particular script it doesn't matter because you're not treating it as interlaced. But with correct handling the field order will matter.

[obaku]

killerteengohan is probably the poster I'm thinking of. Search through his posts https://forum.videohelp.com/search.php?searchid=2094322 as he posts a lot about anime.

Thanks, i will !

JVC SR-V10U (TBC on, Norm, Digital R3=off)

I would strongly urge you to re-record your video in Edit mode. As Norm is just going to smear everything.

I considered using the edit mode at first, but after reading several topics about it i've seen a lot of people advising Norm mode for the hardware noise reduction.
But i re-recorded it anyway, at least i'll have both version to work with. https://slow.pics/c/jKEB3VTd

[jagabo]

Norm mode for the hardware noise reduction.

Hardware noise reduction is not very good -- it will remove detail, and blur the picture. Only use it if you aren't planning to do any filtering in software. Good software can do much better.

[obaku]

You are using a bunch of progressive filters on interlaced video. That's sure to screw things up. Also, your video is TFF but AviSynth assumes it's BFF. In this particular script it doesn't matter because you're not treating it as interlaced. But with correct handling the field order will matter.

Thanks, i'll rework my script (or at least try)

Hardware noise reduction is not very good -- it will remove detail, and blur the picture. Only use it if you aren't planning to do any filtering in software. Good software can do much better.

Should i continue with MCTemporalDenoise or is there a more adapted denoiser for my source ?

[killerteengohan]

DirectShowSource("G:\Record\sjg2.avi", pixel_type="YUY2")
ConvertToYV12(matrix="rec709", chromaresample="point", interlacced=true) # Change to rec601 if SD
ChromaShift(l=0,c=4)
dehalo_alpha(rx=1.2, ry=1.2, darkstr=0, lowsens=0, highsens=100, ss=1.0) # Lower highsens to 50 if you need more white halo reduction
mctemporaldenoise(sigma=1, sharp=false, radius=1, ecrad=1, AGstr=0.6, GFthr=1.00) # Turn off dithering with GFthr=1.00 to save detail from harm
LSFmod(defaults="slow", ss_x=1.00,ss_y=1.0,secure=true,strength=40)
ConvertToYUY2(interlaced=true)

I did not look at any sample, but that should look sharper and have more detail left in tact, guaranteed.

If it's interlaced, you might want to deinterlace the video first, then do all this filtering.

[KarMa]

I considered using the edit mode at first, but after reading several topics about it i've seen a lot of people advising Norm mode for the hardware noise reduction.
But i re-recorded it anyway, at least i'll have both version to work with. https://slow.pics/c/jKEB3VTd

There's a noticeable drop in sharpness and detail. Along with temporal denoising that leads to smearing across neighboring frames.

[jagabo]

Why is there so much ghosting in your beforeprocess.avi file? Whenever something moves you can see several ghosts -- over several frames. And why is the chroma so sharp? VHS chroma is much fuzzier than in this clip. Did you apply some filtering before uploading this clip?

[obaku]

DirectShowSource("G:\Record\sjg2.avi", pixel_type="YUY2")
ConvertToYV12(matrix="rec709", chromaresample="point", interlacced=true) # Change to rec601 if SD
ChromaShift(l=0,c=4)
dehalo_alpha(rx=1.2, ry=1.2, darkstr=0, lowsens=0, highsens=100, ss=1.0) # Lower highsens to 50 if you need more white halo reduction
mctemporaldenoise(sigma=1, sharp=false, radius=1, ecrad=1, AGstr=0.6, GFthr=1.00) # Turn off dithering with GFthr=1.00 to save detail from harm
LSFmod(defaults="slow", ss_x=1.00,ss_y=1.0,secure=true,strength=40)
ConvertToYUY2(interlaced=true)

I did not look at any sample, but that should look sharper and have more detail left in tact, guaranteed.

If it's interlaced, you might want to deinterlace the video first, then do all this filtering.

Thanks, i'll try those adjustments asap !

Why is there so much ghosting in your beforeprocess.avi file? Whenever something moves you can see several ghosts -- over several frames. And why is the chroma so sharp? VHS chroma is much fuzzier than in this clip. Did you apply some filtering before uploading this clip?

I don't know, the clip was totally raw, i didn't apply any filter on it. I thought the ghosting came from bad deinterlacing at first, but it also appears when i separate fields.

[Attachment 53466 - Click to enlarge]
I searched a little and i found this thread : https://forum.videohelp.com/threads/392932-Getting-rid-of-frame-interpolation-in-VHS-captures
if i understand it correctly my record suffers from blend deinterlacing, could it comes from my capture device ?
Or is there a specificity with NTSC-J which could causes this error ?

[jagabo]

The ghosting is not from a blend deinterlace. Some of it may come from a field-blended PAL/NTSC conversion. But some of it appears to be from a strong temporal noise reduction filter -- you can see ghosting for as many as 6 fields. If your VHS deck has a noise reduction filter turn it off (to reduce ghosting). If it has a sharpness filter turn it down (to reduce oversharpening halos). The same for your capture device/software.

[obaku]

The ghosting is not from a blend deinterlace. Some of it may come from a field-blended PAL/NTSC conversion. But some of it appears to be from a strong temporal noise reduction filter -- you can see ghosting for as many as 6 fields. If your VHS deck has a noise reduction filter turn it off (to reduce ghosting). If it has a sharpness filter turn it down (to reduce oversharpening halos). The same for your capture device/software.

Here's a new sample, sjgeditnofilter.avi it got rid of the ghosting.
Like you said, the remaining ghosting probably comes from the conversion of the anime to NTSC standards.
Thanks for helping me on this one !

[jagabo]

That's a lot better. There's much less blending. I did this:

AviSource("sjgeditnofilter.avi") 
AssumeTFF()
Crop(0,0,-16,-0)
ConvertToYV12(interlaced=true)
Interleave(TFM(field=1, pp=0), TFM(field=0, pp=0))
SRestore(frate=23.976) # or maybe 25.0
MergeChroma(last, aWarpSharp2(depth=20))
ChromaShiftSP(x=-3)
Spline36Resize(440,480)
DeHalo_Alpha(rx=2.0, ry=2.5, darkstr=1.3, brightstr=1.3, lowsens=60, highsens=45, ss=1.0)
TemporalDegrain(SAD1=100, SAD2=75, sigma=4)
LSFmod(defaults="slow", ss_x=1.00,ss_y=1.0,secure=true,strength=40)
nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=704, fheight=480)
Sharpen(0.3, 0.0)

There's so little motion in the clip it's not possible to tell if the frame rate should be 23.976 or 25.0. You'll need to check a long moderate speed panning shot to say for sure. I stole killerteengohan's LSFmod, didn't play with the settings.

[real.finder]

for Analog video LGhost and edgesidebleed is always your friend

raw


with edgesidebleed(0,0,0,0.1,0.1)


also SMDegrain() with Dfttest in Spatial mode as mfilter and prefilter help a lot in VHS

[zing269]

There's so little motion in the clip it's not possible to tell if the frame rate should be 23.976 or 25.0.

Japanese anime is almost always done at 24fps and then telecined for distribution. That being said though, those 24 frames often have only 8 or 12 unique images and backgrounds may have even fewer.

[jagabo]

There's so little motion in the clip it's not possible to tell if the frame rate should be 23.976 or 25.0.

Japanese anime is almost always done at 24fps and then telecined for distribution.

Of course. But the presence of blended fields here may indicate this was a PAL->NTSC conversion. In which case the correct frame rate for SRestore would be 25. You could then slow it down with AssumeFPS(24000,1001) if you wanted.

[manono]

Japanese anime is almost always done at 24fps and then telecined for distribution.

Yes, but if the NTSC DVD comes from a PAL 25fps master, then the unblending should be done to 25fps. It might be taken a step further afterwards by slowing it to 24fps (or 23.976fps). I do that myself. To determine to what framerate it should be unblended, a certain amount of movement is required.

Edit: Beaten by jagabo, I got interrupted before posting.

[obaku]

That's a lot better. There's much less blending. I did this:

AviSource("sjgeditnofilter.avi") 
AssumeTFF()
Crop(0,0,-16,-0)
ConvertToYV12(interlaced=true)
Interleave(TFM(field=1, pp=0), TFM(field=0, pp=0))
SRestore(frate=23.976) # or maybe 25.0
MergeChroma(last, aWarpSharp2(depth=20))
ChromaShiftSP(x=-3)
Spline36Resize(440,480)
DeHalo_Alpha(rx=2.0, ry=2.5, darkstr=1.3, brightstr=1.3, lowsens=60, highsens=45, ss=1.0)
TemporalDegrain(SAD1=100, SAD2=75, sigma=4)
LSFmod(defaults="slow", ss_x=1.00,ss_y=1.0,secure=true,strength=40)
nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=704, fheight=480)
Sharpen(0.3, 0.0)

There's so little motion in the clip it's not possible to tell if the frame rate should be 23.976 or 25.0. You'll need to check a long moderate speed panning shot to say for sure. I stole killerteengohan's LSFmod, didn't play with the settings.

for Analog video LGhost and edgesidebleed is always your friend

raw

with edgesidebleed(0,0,0,0.1,0.1)


also SMDegrain() with Dfttest in Spatial mode as mfilter and prefilter help a lot in VHS

Thanks, i took my time to adjust everything and came to a result i really like !
Will provide some clips this week-end

There's so little motion in the clip it's not possible to tell if the frame rate should be 23.976 or 25.0.

Japanese anime is almost always done at 24fps and then telecined for distribution.

Of course. But the presence of blended fields here may indicate this was a PAL->NTSC conversion. In which case the correct frame rate for SRestore would be 25. You could then slow it down with AssumeFPS(24000,1001) if you wanted.

Interleave(TFM(field=1, pp=0), TFM(field=0, pp=0))
SRestore(frate=23.976)

23.976.avi

Interleave(TFM(field=1, pp=0), TFM(field=0, pp=0))
SRestore(frate=25)

25.avi

Couldn't get rid of those blended fields yet. Probably because the base framerate isn't constant.
I'll do more research, it seems to be a recurrent problem