bad blending on FM RF laserdisc capture

[clashradio]

Hello,

I captured my first Domesday Duplicator FM RF project using a NTSC laserdisc. Attempting to IVTC, but it looks like bad frame blending? I've played around with different settings for TFM and TDecimate. None seemed to make any improvements.

The bass players teeth looks like he got punched in the mouth and are in his cheek.

LD Decode outputs an uncompressed v210 or v410 .mov file. I ran the .mov into AVISynth. Since two frames are interlaced and three are not, I IVTC'd. Those results weren't great, so I tried a deinterlace script with QTGMC and SRestore, and it outputted a 25fps file.

Is this one of those NTSC releases that is sourced from a PAL tape? If so is that why this has bad blending? Is there a way to fix the teeth-in-cheek frames?

If it matters, delivery is BD. Below are my two scripts I used. Although I tried various combinations with each script.

v= LSmashVideoSource("side1.mov",format="YUY2")
a= LSmashAudioSource("side1.mov")
AudioDub(v,a)
AssumeTFF()
TFM()
TDecimate()
ColorMatrix(mode="rec.601->rec.709")
nnedi3_rpow2(2,cshift="Spline36Resize",fwidth=960, fheight=720)

v= LSmashVideoSource("side1.mov",format="YUY2")
a= LSmashAudioSource("side1.mov")
AudioDub(v,a)
AssumeTFF()
ColorMatrix(mode="rec.601->rec.709")
QTGMC(preset="faster", matchpreset="faster", matchpreset2="faster", sourcematch=2)
SRestore()
nnedi3_rpow2(2,cshift="Spline36Resize",fwidth=960, fheight=720)

[aramkolt]

Not an expert here, but my guess is the source was not telecine'd appropriately at some stage. Theres' a certain pattern interlaced pattern you can look for when you advance frame by frame (with deinterlacing turned off) where some frames will look progressive and some will look interlaced I believe. Also, your video file and it causes VLC to crash on my Mac, though it usually will play other AVI files uploaded to forums fine, so might be something "uncommon" about your file.

[clashradio]

When I marked in/out the .mov file, I selected Direct stream copy. When I saved the video I was surprised it was avi vs mov. Or should I have selected Quicktime/MOV on Save as type?

[lollo]

The fields already have blended content:

VH_fields.avi

Search the forum about some technique from master jagabo to reduce the defect through blurring

[clashradio]

I added Blur (1.0).Blur(1.0).Blur(1.0). Just made the image too soft. The residual frames are still visible.

I searched the forum & Doom9 for other bad field blending/interlace line posts.

rrats said Aug 4th 2021: "You can blur the video, thus blending the fields, but do this only if any other options don't work".

Quint said Aug 4th 2021: "There are also Functions like De-Bilinear or De-Bicubic. The stripes are caused by Field-Delays (not interlacing) and then resizing in that state. If you are lucky and find the right parameters you can maybe downsize to original size and then fix the field delay. But you need a lot of luck for this, I never managed it".

[Selur]

for general amusement: too sharp (script: https://pastebin.com/YaSndxpK), looking what happens if one throws some machine learning stuff at it. (no, it does not help with the blends, but adds a new look )

[clashradio]

for general amusement: too sharp (script: https://pastebin.com/YaSndxpK), looking what happens if one throws some machine learning stuff at it. (no, it does not help with the blends, but adds a new look )

I see what you mean; your mp4 is too sharp, but I can still see his teeth in different frames.

[jagabo]

I deleted what I did the other day, but I noticed that the cleanest of the frames was usually the last of each string of repeats. After QTGMC() it had a typical repeating pattern like AAABB. So I used SelectEvery(5, 2, 4) to decimate to 23.976 fps. That looked fairly good for the short sample, only a little blending left. But it's likely you'll have to adjust the pattern for longer sequences.

Following that you can use the usual cleaning techniques.

[clashradio]

I deleted what I did the other day, but I noticed that the cleanest of the frames was usually the last of each string of repeats. After QTGMC() it had a typical repeating pattern like AAABB. So I used SelectEvery(5, 2, 4) to decimate to 23.976 fps. That looked fairly good for the short sample, only a little blending left. But it's likely you'll have to adjust the pattern for longer sequences.

Following that you can use the usual cleaning techniques.

Thanks jababo. Adding SelectEvery made an improvement; esp in the chin area.

I don't have a lot of experience with blended frames. Is this a case where you can only get so much water from a rock?

[jagabo]

Is this a case where you can only get so much water from a rock?

I think so. The blending isn't consistent enough that you can deblend by subtracting a small amount of a previous frame.

Here's the script I used:

LWLibavVideoSource("VH.avi") 
AssumeTFF()
QTGMC(preset="fast")
prefetch(8)

Using SMDegrain() may help reduce some of the residual ghosting when there isn't too much, or too complex, motion.

[clashradio]

Thanks for the script jagabo. I see you have prefetch. Couldn't find much info. Is that a multithreading function?

[jagabo]

Yes, prefetch is for multithreading in AviSynth+. If you're using a singled threaded AviSynth discard that line.