Trying to improve movie recorded from OTA to VHS then transfered to DVD

[3d1l]

Hi everyone,

I’m attempting to improve the visual quality of a film that the only copy I could get came from a PAL VHS recording of an standard definition OTA transmission from a RAI broadcast (Italian TV channel), later transferred to DVD. Many years ago I received it from a contact from Italy, but couldn’t play it at the time due to the PAL/NTSC mismatch.

Recently, I extracted the DVD’s VOBs directly to MKV (no recompression) and ran it through HandBrake. As expected, quality was poor (for today standards), SD, analog noise, compression artifacts. I then tried AVSynth+ but still no perceptible improvement.

Motivated by the impressive online restorations I’ve seen of even worse looking material, I decided not to quit and spent weeks configuring a VapourSynth enviroment with all the plugins and filters and building a pipeline. Despite all that, the result is only slightly better. The improvement is so minimal it’s almost imperceptible. The footage still looks soft and degraded, even after all processing.

Here’s the core of what I’m doing:

Deinterlacing with QTGMC

Antialiasing with vsTAAmbk

RGB conversion and upscale with Real-ESRGAN

Color tweaks + neo_f3kdb for debanding

My questions:

Is this kind of minimal improvement expected from such a degraded source?

Am I missing a crucial step in the chain?

Can anything else be done to improve the visual quality, or is this the hard limit?

Any guidance or suggestions would be greatly appreciated.

[lollo]

... a film that the only copy I could get came from a PAL VHS recording of an standard definition OTA transmission from a RAI broadcast
...
Deinterlacing with QTGMC

I suspect then that deinterlacing is not needed, because PAL movies are generally Progressive segmented Frame (even and odd fields are from the same moment in time).

For all the rest, you should post a sample of the video (I suspect upscaling won't improve anything as well).

[Selur]

I agree, a sample is needed to assess whether the file can be improved,.. (upscaling a messed up source will only lead to an even more messed out upscaled version,..)

[Alwyn]

Make a 60 second VOB snippet with AVIDemux. Use the MPEG PS output format option.

[cholla]

Make a 60 second VOB snippet with AVIDemux

To make it clear this would be a snippet from one of the .VOBs on the DVD or a .VOB copied from the DVD.
Not the .MKV .

With enough "action" video to work with.
I only post this so it is not a snippet of opening credits.
Which may be what is at the start of the video.

[Alwyn]

To make it clear this would be a snippet from one of the .VOBs on the DVD or a .VOB copied from the DVD.
Not the .MKV .

My understanding is it shouldn't make any difference because the MKV is just the combined VOBs wrapped in the MKV container.

[cholla]

To make it clear this would be a snippet from one of the .VOBs on the DVD or a .VOB copied from the DVD.
Not the .MKV .

My understanding is it shouldn't make any difference because the MKV is just the combined VOBs wrapped in the MKV container.

I'm not sure if it would make a difference.
I will have to search how to put the .vobs directly from a DVD into a .mkv.
I don't believe I have ever done that.
If I have I was unaware of it.

[Selur]

if it's an unencrypted DVD using mkvtoolnix should work

[Alwyn]

I will have to search how to put the .vobs directly from a DVD into a .mkv.

Use MakeMKV.

To clarify, all we're after is a snippet of the video that needs restoring, which is obviously MPEG 2; whether part of one of the VOBs or part of the ripped MPEG 2 video in the MakeMKV MKV, doesn't matter.

[cholla]

if it's an unencrypted DVD using mkvtoolnix should work

I do not know if the OPs DVD was encrypted or not.
A Google search has several recommendations for MakeMKV.
They say MakeMKV does this.
I have used MakeMKV many times.
I just did not realize it was a direct copy.

I may give mkvtoolnix a try just to see what it does.

[3d1l]

Thanks so much for all your suggestions.

I want to apologize for the delay in my response to your comments.

Sadly, I had an issue with the files and lost most of the notes I took on how I configured HandBrake, and the AviSynth and VapourSynth scripts. I think that information is important for you to evaluate the results. Fortunately, I didn't lose the final videos that I obtained with HandBrake and AviSynth. I never created a full encoded version with Vapoursynth as I was still evaluating the script using VSEdit. I was able to retain one version of the VapourSynth script. There is not problem with the original VOB files because I have them in the DVD so I recreated a new lossless MKV file. I'm preparing a collage of representative images and will post them as soon as possible. In the meantime let me give you more information:

My Computer Specs:

Processor - 1x12th Gen. Intel Core i9-12900H vPro Processor
Memory - 1x32 GB DDR5-4800
Operating System - Windows 11 Pro 64
Hard Drive - 1x 512 GB SSD PCIe
Graphics - 1x NVIDIA RTX A2000 4GB

Video File Info:

The VHS tape was recorded in Standard Play (SP) mode.
Video: MPEG-2 Video 720x576 25fps [Video 1 [Default] (MPEG2, 4:2:2, 0, 720x576, 25.000 fps)]
Audio: AC-3 48000Hz 2.0 chn 256 kbit/s [Audio 1 [Default] (AC-3, 48.0 kHz, 2.0 chn, 256 kbit/s)]
Format profile : Main@Main
Format settings : CustomMatrix
Format settings, BVOP : No
Format settings, Matrix : Custom
Format settings, GOP : N=10
Format settings, picture struc : Frame
Codec ID : V_MPEG2
Codec ID/Info : MPEG 1 or 2 Video
Duration : 1 h 38 min
Bit rate mode : Variable
Bit rate : 4 622 kb/s
Maximum bit rate : 7 500 kb/s
Display aspect ratio : 4:3
Frame rate mode : Constant
Standard : PAL
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Interlaced
Scan order : Top Field First
Compression mode : Lossy
Bits/(Pixel*Frame) : 0.446
Time code of first frame : 00:00:05:05
Time code source : Group of pictures header
GOP, Open/Closed : Closed
Stream size : 3.19 GiB (93%)
Default : Yes
Forced : No

[3d1l]

Thanks so much for all your suggestions.

I want to apologize for the delay in my response to your comments.

Sadly, I had an issue with the files and lost most of the notes I took on how I configured HandBrake, and the AviSynth and VapourSynth scripts. I think that information is important for you to evaluate the results. Fortunately, I didn't lose the final videos that I obtained with HandBrake and AviSynth. I never created a full encoded version with Vapoursynth as I was still evaluating the script using VSEdit. I was able to retain one version of the VapourSynth script. There is not problem with the original VOB files because I have them in the DVD so I recreated a new lossless MKV file. I'm preparing a collage of representative images and will post them as soon as possible. In the meantime let me give you more information:

My Computer Specs:

Processor - 1x12th Gen. Intel Core i9-12900H vPro Processor
Memory - 1x32 GB DDR5-4800
Operating System - Windows 11 Pro 64
Hard Drive - 1x 512 GB SSD PCIe
Graphics - 1x NVIDIA RTX A2000 4GB

Video File Info:

The VHS tape was recorded in Standard Play (SP) mode.
Video: MPEG-2 Video 720x576 25fps [Video 1 [Default] (MPEG2, 4:2:2, 0, 720x576, 25.000 fps)]
Audio: AC-3 48000Hz 2.0 chn 256 kbit/s [Audio 1 [Default] (AC-3, 48.0 kHz, 2.0 chn, 256 kbit/s)]
Format profile : Main@Main
Format settings : CustomMatrix
Format settings, BVOP : No
Format settings, Matrix : Custom
Format settings, GOP : N=10
Format settings, picture struc : Frame
Codec ID : V_MPEG2
Codec ID/Info : MPEG 1 or 2 Video
Duration : 1 h 38 min
Bit rate mode : Variable
Bit rate : 4 622 kb/s
Maximum bit rate : 7 500 kb/s
Display aspect ratio : 4:3
Frame rate mode : Constant
Standard : PAL
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Interlaced
Scan order : Top Field First
Compression mode : Lossy
Bits/(Pixel*Frame) : 0.446
Time code of first frame : 00:00:05:05
Time code source : Group of pictures header
GOP, Open/Closed : Closed
Stream size : 3.19 GiB (93%)
Default : Yes
Forced : No

[3d1l]

I suspect then that deinterlacing is not needed, because PAL movies are generally Progressive segmented Frame (even and odd fields are from the same moment in time).

For all the rest, you should post a sample of the video (I suspect upscaling won't improve anything as well).

Thanks lollo. Actually, the video specs show "Scan type: Interlaced, Top Field First" so deinterlacing is needed in this case. You're right about the sample and upscaling concerns though - working on the images as mentioned in my update above.

I agree, a sample is needed to assess whether the file can be improved,.. (upscaling a messed up source will only lead to an even more messed out upscaled version,..)

Agreed on both points, Selur. Sample images coming as mentioned above, and you're probably right about upscaling making things worse rather than better with this level of degradation.

if it's an unencrypted DVD using mkvtoolnix should work

The DVD wasn't encrypted since it was a homemade VHS-to-DVD transfer. I used FFmpeg to extract the VOBs directly to MKV without recompression.

To make it clear this would be a snippet from one of the .VOBs on the DVD or a .VOB copied from the DVD.
Not the .MKV .

My understanding is it shouldn't make any difference because the MKV is just the combined VOBs wrapped in the MKV container.

You are correct is just a container without recompression. But I'll work on that after posting the image samples... just in case.

[Selur]

"Scan type: Interlaced, Top Field First

Sadly flags are not reliable.

Like mentioned before, you need to share a sample of the clip (not images) if you need further help/suggestions.

[Alwyn]

...

[3d1l]

"Scan type: Interlaced, Top Field First

Sadly flags are not reliable.

Like mentioned before, you need to share a sample of the clip (not images) if you need further help/suggestions.

Got it, I will work on the sample. Your suggestion is to submit a one minute sample but I think that won't capture all the artifacts and defects from all the scenes. So I will try to create a collage in one video file consisting of 3 samples 20 seconds each or even better 6 samples consisting of 10 seconds each.

[Selur]

you can simply share multiple samples,..

[Alwyn]

So I will try to create a collage in one video file consisting of 3 samples 20 seconds each or even better 6 samples consisting of 10 seconds each.

Do not do that. That will recode all the videos and we will be unable to help you to rectify the artifacts.

As Selur says, just give us small samples of each artifact in separate files that have not been recoded.

[3d1l]

you can simply share multiple samples,..

Do not do that. That will recode all the videos and we will be unable to help you to rectify the artifacts.

As Selur says, just give us small samples of each artifact in separate files that have not been recoded.

I followed your instructions and created lots of different video snippets, none of them longer than 1 minute (most of them even shorter), where I tried to capture everything: bright & dark scenes, moving & static scenes, scenes with lots of geometric shapes, scenes with basic & complex colors, scenes with lots of pan & zoom, interior & exterior lighting scenes, scenes with & without faces... you name it.

I think I overdid it just a little, sorry, you ask for few I'm giving you 15 😯 but at least you can't say I can't deliver and I'm a newbie after all 😁

[1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15]

[Selur]

looking at clip1:
For combing handling, I would use TFM().QTGMC(InputType=3)

Cropping seems to be different in the clips.
Luma range is messed up, whites are blown out.
Some clips have Frozen frames, some have chroma bleeding.

Only had time to playing around and applied the same script to all files, maybe it gives some ideas.

Cu Selur

[Sharc]

I didn't check all, but as a simple basic "fix" you may want to try

assumeTFF()
qtgmc().selecteven()
#<crop as needed here>
levels(16,1.0,255,0,235,coring=false)

Encode with SAR 12/11

[3d1l]

selur & Sharc, Thanks! - Working on my Vapoursynth scripts with the suggestions you gave me. Will report on progress later.

[Sharc]

Oops, forgot to add selecteven() in my post#21. Post#21 and attachment edited.

[lollo]

My contribution as well, with sample 002.mpg.

The fields are PsF, but with a phase-shift. I fixed them with TFM(order=1, mode=0, PP=0, slow=2, field=1), no post-processing of residual combed frames of needed. I did not check all frame in details, so if you find some you can adjuste the TFM parameters or use QTGCM as Selur properly suggested:



The capture suffers of line time errors, because probably no TBC has been used:



The capture suffers of clipping in the whites, because probably no level control has been used:



Without addressing other issues (bad levels, chroma noise, chroma bleeding, etc), after the TFM filtering a basic restoration improves a little bit the overall look: https://imgsli.com/Mzg3NzA2

video_org=FFmpegSource2("002.mpg", atrack=-1)

# trimming
	trim_start=800
	trim_end=0
video_org_trim=video_org.trim(trim_start,trim_end)

# cropping 
	crop_left=14	# | rimozione esatta delle bande nere sinistra, sopra, destra e del disturbo sotto	
	crop_top=62	# | 720-(14+12)x576-(62+74)=694x440
	crop_right=12
	crop_bottom=74
video_org_trim_crop=video_org_trim.crop(crop_left,crop_top,-crop_right,-crop_bottom)

# TFM
video_org_tfm=video_org_trim_crop.AssumeTFF().TFM(order=1, mode=0, PP=0, slow=2, field=1) #.TFM(order=1, mode=0, PP=0, slow=2, field=1, display=true)

#eval("video_org_tfm")

#stackhorizontal(\
#subtitle(video_org,"video_org",size=28,align=2),\
#subtitle(video_org_tfm,"video_org_tfm",size=28,align=2)\
#)

### denoising
denoised=video_org_tfm.TemporalDegrain2(degrainTR=3)

### sharpening
sharpened=denoised.LSFmod(defaults="slow")

### adding borders
restored=sharpened.addborders(crop_left,crop_top,crop_right,crop_bottom)

return(restored)

Concerning upscale, I am not a big fan of it, do it only if your TV/Player does a worse job than a software approach.