VHS captures are too dark

[theseeker2]

I'm capturing VHS footage with a JVC S-VHS HR-S3900U hooked up via S-Video to a Panasonic DMR-ES10 for TBC going into an ATI TV Wonder USB 600. I'm using VirtualDub to capture in HuffyYUV with sharpness set to 0, and I learned how to set the crop settings to avoid overscan so that I could adjust the brightness and contrast to eliminate red on the left and right side of the histogram.

Even though I've captured everything within the 16-235 range, my captures seem quite dark. I've already captured hundreds of hours of footage this way, and I haven't found a good way in AviSynth to "fix" it. I'm considering recapturing everything

See below for two sample screengrabs from two different sources, the first from a personal tape, and the second from a commercial tape. These screen captures are from mp4 files that have been deinterlaced with QTGMC but not "corrected" in any way other than running ConvertToYV12 in AviSynth. They do seem dark.

If i do go the recapture route, I need some advice on adjusting the brightness and contrast levels in VirtualDub to brighten things up a bit. If I increase the contrast, then I'll definitely be capturing above 235 and I will see red on the right side of the histogram. Maybe the secret is to accept some red on the right to avoid the captured footage being too dark?

Alternatively, if there's an easy way to "fix" this with AviSynth after the fact, I'd love to know how to do that.

Any input is greatly appreciated! Thank you.


[Attachment 83537 - Click to enlarge]

[Attachment 83538 - Click to enlarge]

[lollo]

Even though I've captured everything within the 16-235 range

Why are you capturing to 16-235? You card is able to capture 16-255, so you should stay in that range. There is always time in post processing to shrink the range to 16-235, when and if needed.

In any case, ColorYUV can manipulate the range at your preferences, playing with gain_y, gain_u, gain_v, off_y, off_u and off_v parameters, just as example.

Post an untouched sample of your capture, not an image from processed and compressed video.

[theseeker2]

Thanks for the tip on the ATI USB 600 being able to capture up to 255, I thought I had to keep it to 235 or under.

I captured some new footage at the full 16-255 range, allowing the histogram in VirtualDub to show red on the right.


[Attachment 83539 - Click to enlarge]

Here's my first ever attempt at running a histogram in AviSynth using this footage.


[Attachment 83540 - Click to enlarge]

Is there anything I should adjust on capture or does this look good? In particular, does that big yellow spike on the right of the histogram mean that I captured beyond the range of the card? Or is that typical for footage that is captured all the way up to 255?

I've also attached a short video sample in case that is useful.

[Sharc]

@theseeker2: You have clipped brights and gaps (missing codes) and spikes in the histogram.
I always discourage newbies from capturing deliberately into the 236...255 range even when the card supports it. It just calls for troubles. There is a 99% chance that you are going to loose details in the brights as soon as you postprocess it with filters which convert the capture from YUV to RGB, using the Rec601 matrix (standard). Result is for examples flat grey skys and similar. The YUV->RGB conversion with clipping may already happen when you import the capture into an NLE. The damage is there.
When you capture into the >235 range you need to process everything in YUV to prevent clipping, or use the studio matrix for conversion.

Even when you want to capture into the superwhites, you should inspect the full video for the setting of the levels, to make sur that even bright scenes are not getting unintentionally clipped at Y=255. Stay in 16.....235, it's just safer.

Edit:
The attached clip shows in the waveform below the picture how badly your brights are clipped at 255. All details are lost.

[lollo]

I captured some new footage at the full 16-255 range, allowing the histogram in VirtualDub to show red on the right.

Too much. Now you have accumulation, and clipping the whites at Y close to 255 (in your case the card is not able to capture up to 255, just a little bit lower).





But the principle is OK, you can use the available range of the card while capturing, to avoid spikes and gaps in the histograms. The gaps in the histogram (highlighted) is what you caused by enlarging the range with the procamp, which is something you should never do.
You should use the procamp just to shrink the range to the one that your card can capture. And later do in postprocessing any change you wish.

Let's say with other words: you check that you have no accumulation at the limits of your card, Y=16 and Y=252/253 in your case. If you do you then shrink the input range with the procamp to stay inside the capturable range. If you see no accumulation you do nothing!

Sharc and myself disagree here somehow: he prefers to always stays in the safe range 16-235 while I prefer to capture in the range allowed by the capture card. The reason is that every procamp manipulation is done at digital level, and introduces gaps (when enlarging) and spikes (when compressing), which is not good, so I try to reduce that at minimum (the gaps are not a serious problem, the spikes can be).

Here an example of mine about accumulation: h.avi



Here a simulation of what happens provided some time ago by master jagabo about spikes and gaps: hist.avi



Sharc's recommendation is somehow safer if you are not a perfectionist. On top of that, as I said, in post processing you may need a RGB conversion or other filtering requiring the levels to be in 16-235 range for a proper manipulation, then you have to shrink anyhow. However, this time is done in post processing, with all the advantages it entails.

To summarize, capture in 16-235 to be safe and have no further manipulation, capture in 16-252 (the range of your card) to extract every bit of quality from your tapes.

[theseeker2]

Thanks lollo and Sharc, I think I'm starting to understand the intricacies of this.

If I do decide to capture up to 252, being careful not to go beyond that, what do I need to do in my AviSynth script in post processing? Here's the script I've been using:

SetFilterMTMode ("QTGMC", 2)
AVISource("I:\testilc.avi")
ConvertToYV12(interlaced=true)
AssumeTFF()
QTGMC(preset="Slower")
Crop(0,0,0,-8)
AddBorders(0,0,0,8)
BilinearResize(960,720)
Prefetch(4)

[lollo]

I would do:

AviSource("I:\testilc.avi")
Crop(8,2,-8,-4)
AssumeTFF()
QTGMC(preset="slow")
AddBorders(0,2,0,4)
nnedi3_rpow2(rfactor=2, nns=4, qual=2, cshift="Spline36Resize", fwidth=1440, fheight=1080)
ConvertToYV12() # if needed for encoding

Before convertion to YV12 you can do a level correction according to your taste and to shink to 16-235 range (with ColorYUV() for instance).

You have Y luma fluctuation in some segment (the histogram moves from left to right to left inside same scene), which is not good, but is probably baked in the tape or an AGC problem in the capture card: a.avi

edit: for instance someting like this (do not trust my parameters, is it just a quick attempt, find your best)

AviSource("I:\testilc.avi")
Crop(8,2,-8,-4)
AssumeTFF()
QTGMC(preset="slow")
ColorYUV(off_y=-25, gain_y=8)
Levels(16,1.0,240,16,235,coring=false, dither=true) # the ColorYUV parameters choice should center the range in 16-235 already, this is just for safety)
AddBorders(0,2,0,4)
nnedi3_rpow2(rfactor=2, nns=4, qual=2, cshift="Spline36Resize", fwidth=1440, fheight=1080)
ConvertToYV12() # if needed for encoding

[jagabo]

In addition to the blown out brights the black level is way too high. This is where a classic histogram is misleading. Here's the same frame lollo posted earlier to show the gaps in the histogram, Histogram(mode="levels"):


[Attachment 83552 - Click to enlarge]

If you crop away the edges of the frame the peak at Y=~17 is much reduced, Crop(16,16,-16,-16):


[Attachment 83553 - Click to enlarge]

Most of the peak is gone. I.e., most of the peak was the black bars at the edges of the frame, not the active picture area. The rest of the peak is from oversharpening halos.

It's hard to tell from that shot but the real black level of the picture is probably about 32 units above Y=16, maybe even more. After Color YUV(off_y=-32) you get:


[Attachment 83554 - Click to enlarge]

That a much better looking image. It's not all washed out and the colors are more saturated, etc.

[jagabo]

Regarding gaps in the histogram: small "errors" like that are of almost no consequence with VHS noise caps. The noise in the VHS cap far overwhelms the quantization gaps in the histogram. Here's an example of an extreme case, using one field of the same frame as above:


[Attachment 83555 - Click to enlarge]

At the top is the original field with about 8 of 256 lines missing, at the bottom the same field with 192 missing lines. Even with that severe damage in the histogram you can hardly see a difference in the image. Mild noise reduction or deinterlacing will pretty much eliminate the gaps in the original video.

[lordsmurf]

to a Panasonic DMR-ES10 for TBC

It'll look like crap, and may drop frames, but test the capture without the ES10.

The ES10/15/etc is not a transparent device, and alters values. Panasonic also has a chromic issue with getting luma accurate on all of their devices going back decades.

ES10/etc is a DVD recorder, not a TBC. It does contain a strong+crippled line TBC, with a non-TBC frame sync. That fact matters, as it exists solely to allow their device to capture video without too many problem. It's not necessarily there "for you" but "for itself". We're trying to tap into that, use it in ways never intended. Sometimes that works great, sometimes not at all.

If you're still getting bad values, then move on to testing another VCR. The 3900 is an EOL consumer S-VHS, and I have seen those badly process.

[dellsam34]

I rarely use histograms, they usually just average the gain of the entire frame, I use my eyes to make sure the most important parts of the frame have the right level, Your captures are indeed dark, You shouldn't capture like that even if there is a possibility to fix in post.

[Sharc]

Regarding gaps in the histogram: small "errors" like that are of almost no consequence with VHS noise caps. The noise in the VHS cap far overwhelms the quantization gaps in the histogram. Here's an example of an extreme case, using one field of the same frame as above:


[Attachment 83555 - Click to enlarge]

At the top is the original field with about 8 of 256 lines missing, at the bottom the same field with 192 missing lines. Even with that severe damage in the histogram you can hardly see a difference in the image. Mild noise reduction or deinterlacing will pretty much eliminate the gaps in the original video.

Thinking about it:
From this experiment with the gaps one may conclude that for "typical noisy" VHS/S-VHS sources a bit depth of 7 bits = 128 steps would actually be sufficient.
Similarly, one may conclude that one gets almost no benefit by exploiting the capture range of a given card to its maximum supported range (i.e. beyond 16....235). In other words one doesn't loose anything noticeable when staying within 16....235 capture range (preventing other troubles). To see the difference one would need a perfect (noise-free) source.
Am I totally wrong?

[lollo]

Thinking about it:
From this experiment with the gaps one may conclude that for "typical noisy" VHS/S-VHS sources a bit depth of 7 bits = 128 steps would actually be sufficient.

Yes, 8 bits capturing is more than enough for such a material.

Similarly, one may conclude that one gets almost no benefit by exploiting the capture range of a given card to its maximum supported range (i.e. beyond 16....235). In other words one doesn't loose anything noticeable when staying within 16....235 capture range (preventing other troubles). To see the difference one would need a perfect (noise-free) source.

I do not completely agree here. When you shrink the histogram to fit to 16-235 (while the card can capture higher) you are not creating gaps in the histogrames, but (useless) spikes (and then data loss somehow); remember our old discussion here: https://forum.videohelp.com/threads/402267-Adjusting-capture-card-levels-not-sure-if-it-s-right

Am I totally wrong?

Never

[theseeker2]

Thanks for all the tips!

I noticed there is some variation in when folks are cropping and adding borders in terms of the QTGMC deinterlace.

My script order is:
deinterlace
crop
add borders

Lollo's order is:
crop
deinterlace
add borders

And I've seen others where the order is:
crop
add borders
deinterlace

Which will result in best quality for a final output of a 60fps mp4 file at 960x720 for YouTube streaming? I don't need to be able to reinterlace as I am keeping the source HuffyYUV AVI files.

[lollo]

I prefere to crop before deinterlacing to prrovide a cleaner frame to the filter, without black borders and head switching noise that may confusion the motion vectors on the edges.

In practice the advantage in term of quality is marginal, so do whatever you prefer. Remember to respect the rules when cropping before deinterlacing, i.e. mod 2 for width and height of YUV interlaced material.

[theseeker2]

Can you all offer some advice on these two image captures? Both from the same tape in the same VCR, but the "old" image has image stabilization enabled on the VCR, and the "new" image has image stabilization disabled on the VCR. I also captured the "new" image at a slightly higher brightness level.

Here's the image slider: https://imgsli.com/MzM2MzY5

I think I can see more detail with the drummer in the back in the "new" capture, but everything also seems "fuzzier" somehow in the "new" capture, not as sharp as the "old" one. Am I just imagining that?

See below for the "old" and "new" images, same as in the image slider:

Old:

[Attachment 84704 - Click to enlarge]

New:

[Attachment 84705 - Click to enlarge]

[Selur]

Am I just imagining that?

no.

[theseeker2]

I figured it out. The reason the "old" capture is sharp and the "new" capture is fuzzy is the sharpness setting on my ATI USB 600 was set to the default of "2" on the "old" capture and set to "0" on the "new" capture. I know there have been many discussions on this forum and others about turning off the sharpness (setting to 0)o on the ATI USB 600, but I can't justify that based on these results. It seems like the default sharpness setting of "2" is best.

[Sharc]

I figured it out. The reason the "old" capture is sharp and the "new" capture is fuzzy is the sharpness setting on my ATI USB 600 was set to the default of "2" on the "old" capture and set to "0" on the "new" capture. I know there have been many discussions on this forum and others about turning off the sharpness (setting to 0)o on the ATI USB 600, but I can't justify that based on these results. It seems like the default sharpness setting of "2" is best.

I can't speak for the ATI USB600, but usually it is better to do any sharpening in post processing and leave the capture devices settings at "neutral". I don't know the "neutral" position of the ATI (0, 2 or whatever ...). In post processing one has usually better control over sharpening artifacts like halos, pixelation, unwanted noise sharpening etc.

[theseeker2]

Thanks Sharc. I guess neutral for the ATI USB 600 is 2, or at least that's how it comes when it's new right out of the box.

[lollo]

On my Hauppauge USB-Live 2 I leave the sharpness setting at medium value (probably corresponding to value 2 in ATI 600USB, but each card is different).

If higher too much artifacts (as pointed out by Sharc); if lower, too soft image.

In addition, if no postprocessing is planned, I found that this setting gives the best "raw" image. If postprocessing is planned, it does not hurt anyhow.

Always experiment with your own material rather than using generic setting, as you did!