AmarecTV Capture Washed out compared to VirtualDub

[BullBuchanan]

Hi,

This is my first time trying to do VHS capture, so please bear with me if I've overlooked something. I'm trying to digitize some rare and out of print folk music recordings for personal use, for some friends and others of interest. I have some modern video editing experience (DaVinci Resolve), but this type of work is all new.

I initially did a capture with VirtualDub2, but ran into audio issues, so I tried VirtualDub 1.9 where I was able to capture, but was still experience audio sync drift that was not consistent throughout the video. I then tried AmaRec which captured with no sync issues. I've been running all captures through StaxRip for QTGMC deinterlacing, scaling etc, but I verified that the saturation/contrast difference is present in the raw captures as shown in the pic.

Both sets of software had contrast and saturation at 128 with all other settings at 0.

Any ideas?


[Attachment 93152 - Click to enlarge]

[Attachment 93153 - Click to enlarge]

Equipment:
VCR: Toshiba D-KVR20
Capture card: Oassuose VHS to Digital Converter
PC: AMD 7700x, 7900XTX, 32GB DDR5 RAM
Staxrip2.52.4
AmaRecTV 3.10
VirtualDub 1.9
VirtualDub2 Build 44282



AmaRec File

Format : AVI
Format/Info : Audio Video Interleave
Format settings : WaveFormatEx
File size : 22.0 GiB
Duration : 56 min 48 s
Overall bit rate : 55.5 Mb/s
Frame rate : 25.000 FPS

Video
ID : 0
Format : YUV
Codec ID : ULY2
Codec ID/Info : Ut Video Lossless Codec
Codec ID/Hint : Ut Video
Duration : 56 min 48 s
Bit rate : 51.7 Mb/s
Width : 720 pixels
Height : 576 pixels
Display aspect ratio : 5:4
Frame rate : 25.000 FPS
Standard : PAL
Color space : YUV
Chroma subsampling : 4:2:2
Compression mode : Lossless
Bits/(Pixel*Frame) : 4.991
Stream size : 20.5 GiB (93%)

Audio
ID : 1
Format : PCM
Format settings : Little / Signed
Codec ID : 1
Duration : 56 min 48 s
Bit rate mode : Constant
Bit rate : 1 536 kb/s
Channel(s) : 2 channels
Sampling rate : 48.0 kHz
Bit depth : 16 bits
Stream size : 624 MiB (3%)
Alignment : Aligned on interleaves

VirtualDub File
Format : AVI
Format/Info : Audio Video Interleave
Format profile : OpenDML
Format settings : WaveFormatEx
File size : 21.8 GiB
Duration : 56 min 46 s
Overall bit rate : 54.9 Mb/s
Frame rate : 25.000 FPS

Video
ID : 0
Format : YUV
Codec ID : ULY2
Codec ID/Info : Ut Video Lossless Codec
Codec ID/Hint : Ut Video
Duration : 56 min 46 s
Bit rate : 51.8 Mb/s
Width : 720 pixels
Height : 576 pixels
Display aspect ratio : 5:4
Frame rate : 25.000 FPS
Standard : PAL
Color space : YUV
Chroma subsampling : 4:2:2
Compression mode : Lossless
Bits/(Pixel*Frame) : 5.001
Stream size : 20.6 GiB (94%)

Audio
ID : 1
Format : PCM
Format settings : Little / Signed
Codec ID : 1
Duration : 56 min 46 s
Bit rate mode : Constant
Bit rate : 3 072 kb/s
Channel(s) : 2 channels
Sampling rate : 96.0 kHz
Bit depth : 16 bits
Stream size : 1.22 GiB (6%)
Alignment : Aligned on interleaves
Interleave, duration : 42 ms (1.04 video frame)

[Alwyn]

That's a big difference in levels. I don't know why; I've done some quick tests with an IOData GV-USB2 and a Hauppauge USBLive 2 and both come out with very similar levels (both with "default" brightness and contrast):

GV-USB2, VDub1911 on the left, Amarec on the right:

[Attachment 93156 - Click to enlarge]

Live2: VDub 1911 on the left, Amarec on the right:

[Attachment 93157 - Click to enlarge]

Perhaps the drivers of your digitiser are a bit dodgy.

If you don't nut it out, you can pretty accurately adjust the Brightness and Contrast levels using the histogram in VDub, then use those values in AmarecTV. I have a blurb on this on my site here.

[Sharc]

Any ideas?

Looks to me like the luma levels got offset (fooled) by the black borders for either case (see attachment). The dark level is actually too high in both cases (especially for the washed out case), lowering the contrast. Can be fixed in post though, or adjusted with proc amp settings for the capturing process.
When evaluating a histogram one should crop off any black borders as these bias the histogram.

Upload snippets (few seconds) of your captures so someone may take a look.

[The_Doman]

This is my first time trying to do VHS capture, so please bear with me if I've overlooked something. I'm trying to digitize some rare and out of print folk music recordings for personal use, for some friends and others of interest. I have some modern video editing experience (DaVinci Resolve), but this type of work is all new.

From your profile and the used Toshiba D-KVR20 I assume you are in the US and using NTSC VHS?
But your capture info shows PAL 25.000 FPS?
Possible here is also a problem?
You should check the capture video format/procamp settings if they are really the same in both capture programs.
When i was capturing PAL60 here using a Sony RDR-AT107 and switching between Virtualdub and AmarecTV, AmarecTV sometimes picked the wrong video NTSC settings and i had to adjust that manually.

[Sharc]

From your profile and the used Toshiba D-KVR20 I assume you are in the US and using NTSC VHS?

The NTSC 7.5 IRE setup pedestal....?

[timtape]

Any ideas?

...Upload snippets (few seconds) of your captures so someone may take a look.

If it's music performance the audio will be important too. Sometimes it's more important than the picture. In the sample please include the unprocessed audio, preferably including a quiet passage and a loud passage.

[BullBuchanan]

This is my first time trying to do VHS capture, so please bear with me if I've overlooked something. I'm trying to digitize some rare and out of print folk music recordings for personal use, for some friends and others of interest. I have some modern video editing experience (DaVinci Resolve), but this type of work is all new.

From your profile and the used Toshiba D-KVR20 I assume you are in the US and using NTSC VHS?
But your capture info shows PAL 25.000 FPS?
Possible here is also a problem?
You should check the capture video format/procamp settings if they are really the same in both capture programs.
When i was capturing PAL60 here using a Sony RDR-AT107 and switching between Virtualdub and AmarecTV, AmarecTV sometimes picked the wrong video NTSC settings and i had to adjust that manually.

So, it looks like this may have been a problem. I'm dealing with a a lot of Irish content, and many of the DVDs I've been digitizing have been in PAL. When I was setting up the capture settings initially, there was no mention on the box or tape of NTSC/PAL and VirtualDub was showing 25 FPS 720x586 when I started playing, and it didn't look off, so I think I incorrectly assumed it was PAL. The video is published by Iowa Public Television, which made it seem weird it wouldn't be NTSC, but that's just my ignorance.

I'm re-recording in virtualdub now with 29.97 and 720x480 and I'll see what happens.

Regarding the video color, even the VirtualDub video preview looks fully saturated liek the final output while the AmaRecTV version is washed out. If I'm not able to get the audio sync issues resolved with virtualDub, I'll see if I can experiment more with AmaRec settings.

[BullBuchanan]

Any ideas?

Looks to me like the luma levels got offset (fooled) by the black borders for either case (see attachment). The dark level is actually too high in both cases (especially for the washed out case), lowering the contrast. Can be fixed in post though, or adjusted with proc amp settings for the capturing process.
When evaluating a histogram one should crop off any black borders as these bias the histogram.

Upload snippets (few seconds) of your captures so someone may take a look.

Those black borders were coming just from VLC I think, but I'm re-capturing now that I've been informed I was incorrectly capturing as PAL. I'll see if that makes any difference.

[BullBuchanan]

After re-capturing a few times due to some errors, I finally got a good capture with AmaRec. I tried re-capturing with VirtualDub, but still had audio sync issues I couldn't resolve.

Color looks correct.

[Attachment 93167 - Click to enlarge]

I processed the Amarec file in StaxRip and resized to 960x720, which I now realize wasn't a proper scale for the aspect ratio. It seems most people say the modern approach is to crop down to 704w and then scale that up to 1440x 1080. Is that the latest approach? do the missing 32 pixels get stretched or do they become a black bar? Is looks like there's a bit of black bar already on the left side of the file. Any downside to just scaling up as is?

As far as upscaling quality, I've read on this thread that AVISynth can be a replacement for topaz AI for VHS quality material. Does anyone know of a good guide on setting this up? StaxRip seems to have an AVISynth integration and De-noise filters available.

https://www.digitalfaq.com/forum/video-restore/12401-upscaling-ntsc-vhs.html

And does anyone have a recommendation for final compression? My 57 minute 720p file is about 60GB, which is about the size of a remuxed 4k blurray. We certainly aren't dealing with the same amount of data here. Is there a lossless compression that will take this down meaningfully in size, like FLAC does for WAV in audio?

[Sharc]

Typical basic steps for the video processing are:
1. Capture lossless as 720x480, 4:2:2, interlaced
2. Deinterlace (QTGMC)
3. Crop to 704x480
(4. Denoise and apply any other filters as needed)
5. Upscale to any 4:3 ratio, like 1440x1080 (square pixels)
6. Encode with x264 (for example) using 1-pass encoding with CRF of your choice, e.g. CRF=18)

(Instead of 5. upscaling you could leave it anamorphic 704x480 and flag it accordingly as SAR=10/11 in step 6)

You can find plenty of posts about this process in the forum.

[BullBuchanan]

Typical steps are:
1. Capture lossless as 720x480, 4:2:2, interlaced
2. Deinterlace (QTGMC)
3. Crop to 704x480
(4. Denoise and apply any other filters as needed)
5. Upscale to any 4:3 ratio, like 1440x1080 (square pixels)
(6. Encode with x264 using 1-pass encoding with CRF of your choice, e.g. CRF=18)

Thanks. Looks like I'm on the right track. For cropping, would I take it all off the left in this case or equal from all for edges? Now that you're left with 704 where does the extra 16 come from to get to 1440 instead of 1408

[Sharc]

Thanks. Looks like I'm on the right track. For cropping, would I take it all off the left in this case or equal from all for edges?

You take a total of 16 off, like 8 left+8 right, or 12 left + 4 right etc. Look at the raw capture (should be 720x480) and crop keeping in mind to remove black bars on either side.

Now that you're left with 704 where does the extra 16 come from to get to 1440 instead of 1408

Your capture is anamorphic (non-square pixels) and the 704 get "stretched" to 1440 by the resizing, which converts the source's non-square pixels to 1440x1080 square pixels, good for viewing as 4:3.

[VWestlife]

3. Crop to 704x480

Why? I've done plenty of VHS captures where the image goes almost all the way to the edges of a 720x480 frame. On Video8/Hi8 footage, it always does. There are no black edges to crop off. So why lose part of the video frame? I just de-interlace and upscale, with no cropping.

And even if there is a slight border area, so what? Virtually nobody is going to be watching it on a native 4:3 display anyway. They're going to be windowboxing it to 16:9 anyway, which adds even more of a black border on the sides. So cropping off the black edges just to add even more black edges upon playback is a pointless waste of time.

[Alwyn]

If you chop off the 16 pixels and have black bars remaining (might be only on one side) and you don't want any bars at all eg for a YT video, just crop all sides in a 4:3 ratio: if you take an extra 6 off one side, you need to take 4 off the vertical (top of bottom, doesn't matter, could be 2 and 2). But make sure you crop in even numbers. This technique is particularly relevant if you want to remove the ugly head-switching noise at the bottom of most VHS captures.

If you want to absolutely minimise cropping and want to preserve the correct aspect ratio and are happy to live with side bars, then crop only what you need then encode to a specific SAR as Sharc notes in post #10 above.

[The_Doman]

Side borders are mostly not a real problem today, but masking them with pure black always gives a cleaner/more compressible image.
For head switching removal I like to crop and add black top/bottom borders to recenter the image again.
But often there are other annoying disturbances you want to crop and/or mask.

Rate my Setup - Capturing Digital8 and Analog

If you really want more "perfection" you could center the image after cropping by adding top and bottom borders.
Really useful if you need to crop a lot extra VHS head switching noise.
But be careful, with YUV you need to use MOD 4 sizes when cropping/centering vertically!

For example you could do something like this.
This is from DV pass-trough, I added a 2 pixel border, and the pictures are (of course) not AR resized!
I let the player(s) resize and, depending on the destination, de-interlace too.


[Attachment 82384 - Click to enlarge]


[Attachment 82385 - Click to enlarge]

[Sharc]

3. Crop to 704x480

Why? I've done plenty of VHS captures where the image goes almost all the way to the edges of a 720x480 frame.

Yes, but nevertheless the pixel aspect ratio for 4:3 NTSC VHS captures is in very good approximation 10/11=0.91 non-square pixels, because the luma sampling rate has been standardized in Rec.601 to 13.5MHz, and any reasonable capture card complies with this. Hence the 4:3 picture is contained within 704x480 1). If your captured active picture goes beyond 704, for example say 720 (overscan) you should actually upscale your 720x480 frame to 1473x1080 for an undistorted picture - which is not exactly 4:3. If you just upscale it to 1440x1080 (=4:3) your picture gets slightly horizontally squashed, means circles become slight ovals. Anyway, the error is within 2.4%, so casual viewers probably won't even notice it. No reason for sleepless nights.
(Also, the transitions from the active picture to the black borders is often fuzzy, so cropping does no harm).

1) If your card captures 486 pixels in height rather than 480 the width of the active picture would become 711 - which would leave 9 pixels only for black borders. Some drivers capture 720x486 and discard (mask) 6 pixels vertically to obtain 480, but keep the 711 horizontal for the active picture. So your milage may vary here according to your capture driver. NTSC is a bit more tricky here than PAL IMO. Whatever applies, the PAR (Pixel Aspect Ratio) of ~10/11 is valid in any case and resizing should be based on this PAR (aka SAR in x264) to keep the shapes intact (circle -> circle).

This has been discussed in much more detail many times in the past. Just lazy to do the search.
It's all a bit confusing in view of an error of max. 2.4% .....

[rgr]

Side borders are mostly not a real problem today, but masking them with pure black always gives a cleaner/more compressible image.

Masking made sense back when you burned that kind of material onto DVD; nowadays, you simply 1. crop it, 2. save with correct PAR.

[BullBuchanan]

So I got this one processed, though I have run into a couple of new issues. The first one seems to be a 5 frame mismatch that wasn't showing up when I was processing these to uncompressed .mov, but do seem to be showing up now. I ran it twice, disabling the "abort" flag from the menu as recommended elsewhere on the forum. The file plays and doesn't seem to have any issues, it this anything to be worried about?

Settings and error here

[Attachment 93171 - Click to enlarge]

[Attachment 93174 - Click to enlarge]

What's a little more annoying is that my "upscale" and "denoise" process seems to have degraded the image quality pretty noticeably. Note, both of these screenshots were taken on a 42" 4k OLED monitor, so anything but pristine images/video looks pretty awful. The original ULY2 AVI is on the left of both images, while the x264 .MOV is on the right.

The first image shows a lot more noise on the grey-blue background, the guitar strap and around the singer's face.

[Attachment 93172 - Click to enlarge]

On the second screenshot there's more noise throughout the entirety of the singer's silhouette. Is there enough information to say if this is compression related or a result of the denoise/scaling filters?

[Attachment 93173 - Click to enlarge]

[Alwyn]

The original ULY2 AVI is on the left of both images, while the x264 .MOV is on the right.

Are you sure you've got the left and right right? For the guy in the blue shirt, the left image is miles better: it's the correct aspect ratio, there's no chroma noise (compare the shirts) and it looks quite nice (eyeballing it, the levels are a bit washed out, methinks).

Re awful on big TVs, I find SD video generally looks better because you are, relatively, much further away than your 27" monitor 2 ft in front of your face.

[BullBuchanan]

The original ULY2 AVI is on the left of both images, while the x264 .MOV is on the right.

Are you sure you've got the left and right right? For the guy in the blue shirt, the left image is miles better: it's the correct aspect ratio, there's no chroma noise (compare the shirts) and it looks quite nice (eyeballing it, the levels are a bit washed out, methinks).

Re awful on big TVs, I find SD video generally looks better because you are, relatively, much further away than your 27" monitor 2 ft in front of your face.

Uh, you're totally right. I have no idea how I mixed them up. I guess all is good then. Thanks.

[Sharc]

Uh, you're totally right. I have no idea how I mixed them up. I guess all is good then.....

Yep, basically. As a possible improvement you may want to add some chroma shift and chroma sharpening to your processing. Maybe StaxRip offers such options, I don't know.

Left: your current version. Right: same with chroma shifted 8 pixels to the right (means better aligned to the luma) and chroma sharpened.

[The_Doman]

Masking made sense back when you burned that kind of material onto DVD; nowadays, you simply 1. crop it, 2. save with correct PAR.

Yes of course all true.
But I quite often also make DVD versions because it is still a nice match for VHS and very convenient for people to play in their DVD/Blu-ray players.
Also most TV's do a pretty good job de-interlacing//up-scaling legacy SD.
And it is just interesting to do.

[VWestlife]

Yes, but nevertheless the pixel aspect ratio for 4:3 NTSC VHS captures is in very good approximation 10/11=0.91 non-square pixels, because the luma sampling rate has been standardized in Rec.601 to 13.5MHz, and any reasonable capture card complies with this.

You think manufacturers actually cared about what the official ITU rules said? Sony used NTSC-J on their American-market equipment instead of the official proper NTSC, and nobody ever cared.

Hence the 4:3 picture is contained within 704x480

Then every standard-definition camcorder from Sony, JVC, or Canon I've ever used must be illegal because they all record 4:3 or 16:9 video to a full edge-to-edge 720x480 frame. Only Panasonic was actually stupid enough to follow the 704x480 recommendation, and they eventually gave up on it and joined the rest of the industry by switching to 720x480 in 2010.

Anyway, the error is within 2.4%, so casual viewers probably won't even notice it. No reason for sleepless nights.

Hence the reason why manufacturers didn't care and recorded 4:3 video to a full edge-to-edge 720x480 frame, regardless of what the official rules said. (Just like how Europe's AC mains voltage is "officially standardized" at 230 volts, but they made the tolerance wide enough that it still varies from 220 to almost 250 volts depending on which country you go to.)

[Alwyn]

The thread's about a VHS capture, not a camcorder recording.

[Sharc]

You think manufacturers actually cared about what the official ITU rules said?

Yes they did and were happy that standards evolved, and current capture cards still stick to the 13.5MHz luma sampling rates which dictates the Pixel Aspect Ratio. Or take some old 4:3 DVDs and see its black side borders. But yes, over time in the digital age this got softened (driven by the DVD industry) and the ITU PAR was replaced by the so called "generic" PAR, or eventually using more convenient square pixels like for blu-rays, for example.
It has all been dicussed to death, and I don't want to argue. So let people continue to ask why they get some black bars with their captures and what to do with these etc. And there are other issues like color space conversion when upscaling from SD to HD (601 vs 709), or level range issues like limited vs full range etc. Just tell people they shouldn't care, and wipe everything off the table with "tolerances" or similar, it is much simpler, I agree. Keep things simple, but not any simpler.
Hope you don't mind my sarcasm

[lollo]

Sharc is right about this subject, the spec is clear. Whatever camcorder and capture card producers do, is not technically correct when they do not adhere to the specifications.

PAL specification attached , the law of physics cannot be changed (13.5MHz, 52usec per line length, 74.074ns -> 702/704 real pixel)

[BullBuchanan]

Uh, you're totally right. I have no idea how I mixed them up. I guess all is good then.....

Yep, basically. As a possible improvement you may want to add some chroma shift and chroma sharpening to your processing. Maybe StaxRip offers such options, I don't know.

Left: your current version. Right: same with chroma shifted 8 pixels to the right (means better aligned to the luma) and chroma sharpened.

Thanks! How did you calculate it needed to be 8 pixels to the right?

[lollo]

Thanks! How did you calculate it needed to be 8 pixels to the right?

There is no calculation to do, just experiment with the numbers and find the best outcomes. PAL generally has intrinsic 2 pixels shift (which adds for every copy-generation); it can be more.

[Sharc]

How did you calculate it needed to be 8 pixels to the right?

As @lollo said it is basically found by trial and error. To start with something look at the discolored area between the guy's skin and the hair (marked with a circle). From this you can guess that the shift is probably somewhere around 10 pixels. I used Avisynth and AvsPmod as convenient tools.

[rgr]

So I got this one processed, though I have run into a couple of new issues.

1. ConvertToYUV420 <-- Use this at the very end. Filtering should be done to maintain quality on the source material, which is probably YUV422. Preferably still at 10 or 16 bits.
2. Resize ✓ Misc Spline36 Resize <-- Think about whether you need upscaling. This usually does not improve quality, especially basic ones. So - NEVER UPSCALE SD.

Besides, I advise against AmarecTV - it loses frames.