request: analog NTSC capture samples

[Wilbert]

As you know we are making an analog capture guide. Since we are all PAL we want to have a bunch of NTSC samples. More concrete, we would like to have the following:

Captures (of the same clip) at three resolutions: 720x480, 704x480 and 640x480 (not processed). Look for a common frame (the image mustn't contain any black by itself), and make three images (for every resolution one) PNG is fine. Equally important, mention what capture device you are using. We want to know what captured devices perform internal scaling, and at what resolution.

Unrelated do the above. I also would like to have some NTSC captures (XviD, not processed). 25-50 frames is fine (as long as you keep it below 1MB).

If you don't have a webpage, sent it to w.j.dijkhof@tue.nl. Thanks for your coorporation!

[BaronVlad]

Any reaction ?

Your help is highly appreciated.

Thanks:)

[jggimi]

I'll be able to give you a sample, but not for another day or two...

[jggimi]

Wilbert,

I was unable to get a repetitive sample -- due to rearrangement of the family's S-Video capable VHS tape deck. It's impossible to extract it from the living room, if I want to stay married, or have kids who still love me. We have other VHS decks in the house but they are composite decks that are so horrendous in quality I wouldn't want to use 'em.

I was able to help you with your second quest -- a few seconds of NTSC samples. I have both Telecined as well as standard interlaced content. Hunting for usable interlaced video was a challenge. (Can you believe they waste film on dreck like "Will & Grace" -- It's Telecined!)I captured at 720x480 (1.5:1) YUY2, in HuffYUV with 172k PCM, and captured with an AverTV BT878 card, using the most recent - v5.3.6-1 - BTWinCap drivers with AVI_IO.

I'll make you unresized/unfiltered XviDs and e-mail them to you at the above address.

[Wilbert]

I'll make you unresized/unfiltered XviDs and e-mail them to you at the above address.
Thanks for your clips!

I was unable to get a repetitive sample (...)
Anyone else up to the challenge? There must be more people who capture NTSC :)

[jggimi]

Oooh. I just had an epiphany:

----
My Laptop has S-video out. It would be really easy to take a "still" -- If you like, I can create a 1024x768 test pattern on the monitor, capture, and send you stills of various capture formats.
----

Come to think of it, I have a DVD with some audio and video tests on it, and I think it has a circle for aspect ratio examination. The tests were on a regular R1 Hollywood film, and the tests were there as extras. The tests were either from THX or Dolby Labs...but I just can't remember which movie it is!

[Wilbert]

My Laptop has S-video out. It would be really easy to take a "still" -- If you like, I can create a 1024x768 test pattern on the monitor, capture, and send you stills of various capture formats.
That would be nice!

[jggimi]

I've e-mailed .png files.

[FredThompson]

Well, this is interesting. I was just working on a similar how-to. How is this different than the guide at: http://www.doom9.org/index.html?/capture/start.html ?

Can send you some All-in-Wonder Radeon captures in the next couple of days.

FWIW, the best "test" samples will have subtle lights and almost-blacks. Sand, concrete, water, (I see) white people, all have a tendency to get washed out.

I've done a huge number of tests and tried all kinds of calibration including the various "theater" calibrations (THX, that kind of thing.) Basically, I've come to the conclusion it is very hard to get a perfect picture with the AiW line. My guess is they don't have linear behavior. I might be wrong.

Also, it seems to me ATI has changed their registry names and I've not been able to figure out exactly where the defaults are saved. It would be very helpful to set the default capture values once you're found settings you like...

Are you only interested in the performance of cards or did you have a larger scope? For example, there's a lot of bad advice where people suggest capturing at the resolution of the end product. Big mistake if you care about quality.

[jggimi]

Well, what I sent to Wilbert was highly compressed ... 2000kbps Xvid. So he won't be able to see the detail level, if any, that was captured in lossless form.

I did send him 4 .png's from a couple of test pattern captures -- he could at least see the stills, though, if nothing else. Unfortunately, those were of color graphs and circles (THX), for examining color convergance and aspect ratio, not detail, unfortunately..

[FredThompson]

I have the THX Optimizer home edition DVD. If you'd like, I could extract frames and send them as PNG files. Here's info:

http://www.thx.com/mod/techLib/index.html

Granted, that won't directly help with testing but, I suppose, a full D1-formatted disc could be made to be played on a hardware player into your sampling device. That would give you at least somewhat of a consistent baseline.

Actually, it's something I'd thought about quite a while ago when I was still trying to find an optimal way to tune the AiW cards. The project never left the planning stage because I moved on to pass-through DV and hardware MPEG-2 encoding. Still, it might be a very helpful thing for folks, a tuning process for both analog capture and output.

The idea was to first calibrate capture using a hardware digital video source. Then output could be calibrated which is dependent on the capture for its testing. I'm assuming a table of baseline calibrations could be generated so somebody with capture card X would know what settings are recommended. The frosting on this cake was to be automatic routines to set default values to the calibrated ones, not the shipped defaults.

(Edit: Accurately setting defaults on the sampling cards should NOT be done by visual analysis of the computer's monitor. Using the monitor adds another layer of color management and it's too confusing. The idea is to calibrate for capture, tuning a computer monitor is a different task and varies dramatically. We can't trust any monitor to accurately depict what the card has sampled. See www.aim-dtp.net http://www.displaymate.com and PC Magazine Oct 4, 1999 (can't seem to find it on the web. I have the article and the sample DisplayMate screens they used) for more information.)

The goal I had was an automated analysis system that would generate the samples it needed. Sort of like the way Gordian Knot "rides on top of" a bunch of other applications.

Here are my design ideas from when the project stopped:

I have an idea about how to standardize calibration of analog grabs using VirtualDub and AviSynth. I'd appreciate your feedback on the basic ideas and implementation. I've just been thinking about the process and haven't had time to look into the programatic issues (scripts, filters, etc.)

The end result should be a

I'm toying with the idea of using Python to automate the entire process as much as possible but that might not happen. It's probably most beenficial to get the process normalized and released to the public and THEN to create automation.

What I'm hoping to do is generate "best fit" settings for analog grabs taking into account the limitations of consumer-level video grab cards. I'm using an ATI All-in-Wonder Radeon but this concept would also apply to the various Brooktree cards.

Remember, this is to account for hardware performance, not any particular analog source.

step 1:

Make a sample grab using VirtualDub to set the blackest point. Assuming the source is videotape, there shoul dbe some overscan. We don't need a lot of source material, 1 second will give us more than 20 sample images. So we record 1 second of source at 640x480 uncompressed.

This 1 seocnd clip is run through analysis that looks at a few points in the overscan areas to read "black" sample values to give the default value for the histogram during capture.

Step 2:

AviSynth script to generate a standardized color bar set. Again a second or two should be enough.

Step 3:

Write the color bar file to VCD or SVCD.

Step 4:

Use VirtualDub to capture the color bar sample played from a DVD player.

Step 5:

AviSynth script takes pixel samples from each of, say, 20 frames of the capture and calculates average sample values for each color swatch.

Step 6:

These values are analyzed to generate proper "correction" values for brightness, contrast, saturation and hue values (hardware, during capture.)

Step 7:

Steps 4-6 are repeated until a "best fit" is achieved.

At this point, the capture settings are assumed to be good.

Step 8:

Perhaps a fine tuning of this process using Donald Graft's hue/saturation/intensity VirtualDub filter as a post-capture process. This plug-in is more adjustable than the basic 4 hardware adjustments.

Step 9:

The AviSynth-generated sample file is exported to video tape and the sampled for analysis. This is a similar analysis but with the intention of fine-tuning analog ouput characteristics of the video card. This would be a group of settings for Donald Graft's hue/sat/bright filter used to tune a file for analog output.

[FredThompson]

Scratch part of what I typed earlier. Here's a completely free test screen generator:

http://www.synthetic-ap.com/products/tpm/index.html

I don't have the time right now but it should be easy enough for someone to make a test VCD/SVCD/DVDR set that is easily downloaded and can be used to test capture performance.

Pretty important to have standardized images if you want to compare capture cards. If it's really important, I'll shoot some DV of the type of real-world source that capture cards typically choke on. Just let me know.

[Wilbert]

@FredThompson

If you want to help:

Captures (of the same clip) at three resolutions: 720x480, 704x480 and 640x480 (not processed). Look for a common frame (the image mustn't contain any black by itself), and make three images (for every resolution one) PNG is fine. Equally important, mention what capture device you are using. We want to know what captured devices perform internal scaling, and at what resolution.
Picture doesn't need to be a circle (unless you can't get a common frame).

[FredThompson]

I must be missing something. How do screen grabs of variable source help?

[Wilbert]

Well, 720x480 contains the full image (besides the six vertical cropped scanlines). It is either stretched to 720, or 9 pixels overscan is added (711+9=720). 704x480 should be missing 7 pixels. 640x480 is either scaled down to 640 (648->640) (in that case it contains the full image), or 8 pixels are cropped.

So, if you have a common frame, you can see in which case there is internal scaling.

[FredThompson]

Oh, I hadn't thought of that.

I always do full D1 then resize as the last step before recompression.

My card is AiW Radeon. Do you still need samples from that?

Guess you'll need the driver version, too, huh? Resizing from native to the requested resolution might be a combination of hardware and driver.

[Wilbert]

My card is AiW Radeon. Do you still need samples from that?
Yes, that would be great!

[FredThompson]

ok, give me a few days and I can do it.

The North American version of this card WILL capture PAL source if it's delivered over composite or s-video cables.

The following are due to arrive on Monday and Tuesday:

JVC HR-J72MS multi-format 6-head VHS deck
ADS Instant USB 2.0
Italian PAL VHS tape

For source, I can do use the following:

NTSC VHS
NTSC S-VHS
NTSC DV
PAL VHS (Italian and German sources, different decks)

What is most helpful? Would you really want all four variants at each of the resolutions?

[Wilbert]

NTSC VHS
NTSC S-VHS
Are fine. Btw, is there's something difference when capping from VHS or S-VHS (assuming it is an analog tv recording)? I guess so, but I don't know anything about s-vhs.

[FredThompson]

ntsc vhs is 200 lines, s-vhs is 400.

[FredThompson]

Is your guide only going to cover capture or are you planning to also talk about how to analyze the captures to see what types of sources were used so a captured interlaced source is properly converted to progressive?

Maybe "properly" isn't the correct phrase. More than one way to reach an acceptable result.

Also, what about noise removal? Logos?

There seem to be two levels of experience and not that much documentation that bridges them. For example, if you poke around you'll see discussion of chroma shift (newbie says, "huh?"), dust filters, temporal filters, etc.

To make an analogy with flatbed scanners: knowing how to scan an image is one thing, knowing how to "clean" it is another. http://www.scantips.com shows what I mean.

I've just realized there are "masters" and "newbies" but there don't seem to be a lot of "apprenticeship" programs.

[Wilbert]

It's not only my guide: http://forum.doom9.org/showthread.php?s=&threadid=50438

Is your guide only going to cover capture or are you planning to also talk about how to analyze the captures to see what types of sources were used so a captured interlaced source is properly converted to progressive?
Yes. I'm satisfied with the AviSynth part. The VDub part is not complete yet.

Also, what about noise removal?
It is/was not the intention to include a comparison with different denoisers. But I mentioned which smoother you could use, and what are the important points to consider for smoothing.

Logos?
Not yet, will be added in a next version.

if you poke around you'll see discussion of chroma shift
Must also be included in a next version.

[FredThompson]

Uh...I started to post this as its own thread then changed my mind.

Please let me know how I can best take this idea and turn it into helpful info for you. The next thing I was thinking about was that it would be a good idea to create some test samples (SVCD, VCD, DVD) that are really small so other people could easily download them and use to make their own captures to expand the comparison. I'm assuming you're going to want some kind of table to show performance so folks will have an idea how things really perform. Also, just realized it's going to take a while to get good torture samples. I remember what the AiW would choke on, just not sure where I've got good samples. I know where a VHS tape is that has a caucasian guy in white trudging through sand. That's a good one ;) Gonna find something with light reflecting off water.

Anyhoo, please take a look at these incomplete thoughts and direct me how best to provide you the samples you've requested:

I've just received an ADS Instant DVD 2.0, still in the wrapper ;)

Question is, how best to post a review?

It will take about a week to full test this thing, maybe two. If it's going to be done, it should be done properly. Or, at least, tot he best of my ability and sanity at the time ;)

I've got these formats playback formats available:

NTSC VHS
NTSC S-VHS
NTSC DV (via composite)
NTSC DV (via S-video)
PAL VHS

The following capture hardware is available:

ADS Instant DVD 1.0
ADS Instant DVD 2.0
Canon ZR65MC NTSC DV camcorder (cannot accept PAL input)
ATI All-in-Wonder Radeon

The following playback hardware is available:

Canon ZR65MC NTSC DV camcorder
JVC HR-S7600 S-VHS NTSC VCR
JVC (durn it, forgot model, it arrives today) VHS multi-format VCR

Source will probably be a combination of DV shot out my window for almost static, high complexity (trees, etc.), something captured to DV from my satellite dish (DirecTV), and the Snell&Wilcox torture sample I've picked up.

Question is, where should I submit this? there will probably be a lot of screen grabs

[Wilbert]

@jggimi and all,

I was wondering. What do you do if you are capping telecined material, but you have frame drops. Do you correct for this somehow?

@FredThompson,

Question is, how best to post a review?
PNG is fine.


NTSC VHS
NTSC S-VHS
NTSC DV (via composite)
NTSC DV (via S-video)
PAL VHS
The guide is about analog capturing, thus
NTSC VHS @ 704x480, 720x480 and 640x480 (common frame)
PAL VHS @ 704x576, 720x576 and 768x576 (common frame)
is fine.

Question is, where should I submit this? there will probably be a lot of screen grabs
Mail them to w.j.dijkhof@tue.nl (attachments can be maximal 1MB).

For you other suggestion:

The next thing I was thinking about was that it would be a good idea to create some test samples (SVCD, VCD, DVD) that are really small so other people could easily download them and use to make their own captures to expand the comparison. I'm assuming you're going to want some kind of table to show performance so folks will have an idea how things really perform.
That would be a very good idea! We could take various sources (for example the telecined and trully interlaced source which jggimi mailed, similar test captures for PAL), encode them to SVCD, VCD and DVD (only resizing, without any further post processing). Then people could capture them, and compare their captures. Is this what you have in mind?

[jggimi]

...What do you do if you are capping telecined material, but you have frame drops... In my small experience with drops, the resulting video has duplicated frames, and the framerate is unchanged. IVTC processing (I usually use Decomb) continues as normal.

e.g: If the video source has ten frames:

1-2-3-4-5-6-7-8-9-10

If frame 5 is dropped, my capture contains:

1-2-3-4-4-6-7-8-9-10

----

Hope this helped!

[Wilbert]

Assuming that 3,4, 8,9, etc. are interlaced. We get:

1p-2p-3i-4i-4i-6p-7p-8i-9i-10p

Three out of the first five frames are interlaced. Which means (I guess) that there is one interlaced frame left after ivtc?

[jggimi]

Yes, the frames may occur that way.

My understanding of the Decomb filters is that, as long as you're using Telecide with post=true (the default), any interlaced frame that slips through the field matching decision process with artifacts will get deinterlaced.

With Telecide, I believe that even if the frame is a scene transition (where one field has nothing to do with another), it will rebuild frames appropriately.

I'll ask Neruon2 to review this post, and comment and or correct me if I am inaccurate or misleading..

[FredThompson]

Originally posted by Wilbert
That would be a very good idea! We could take various sources (for example the telecined and trully interlaced source which jggimi mailed, similar test captures for PAL), encode them to SVCD, VCD and DVD (only resizing, without any further post processing). Then people could capture them, and compare their captures. Is this what you have in mind? [/B]

Yes. My original idea was to use AviSynth to create color swatches for calibrating capture and playback adjustment. That should compress quick well.

Should have some other stuff too. Like I've mentioned, sand, sun on water, white concrete, etc. are hard to capture properly.

Hadn't thought of other types of source.

Clips would not need to be long, maybe 10 seconds at most with looping so they play 30 seconds or so then back to the menu.

Probably best to really plan this out before releasing it.

[Guest]

Assuming a drop results in a duplication of the previous frame, there are two possibilities, depending on which rame is dropped. Either Telecide() will emit an extra progressive frame, or it will emit two combed frames. As jggimi states, if postprocessing is enabled, the combed frames will be deinterlaced. Of course, you will have a little stutter due to the drop.

If pattern guidance is enabled, you will fall out of pattern but it will recover pattern usually within one or two cycles. Being out of pattern just means you don't get the benefit of being able to predict the next match, it will still apply blind field matching.

[jggimi]

Thank you!

[Wilbert]

@jggimi & neuron2
Thanks, for your explanations.

@FredThompson,

Yes. My original idea was to use AviSynth to create color swatches for calibrating capture and playback adjustment. That should compress quick well.

Should have some other stuff too. Like I've mentioned, sand, sun on water, white concrete, etc. are hard to capture properly.
This sounds nice, but I will come back to you later about this.

Separately from this, I would like to have the following:

NTSC VHS @ 704x480, 720x480 and 640x480 (one common frame)
PAL VHS @ 704x576, 720x576 and 768x576 (one common frame)

edit: capped with your ATI card!

[FredThompson]

Yes, I know.

I'll make you the samples but it will have to wait a couple of days. That computer is down and I don't have time to rebuild it until the weekend.

[dopeskills]

Would NTSC from a Beta camcorder (via composite) be usefull?

[Wilbert]

Would NTSC from a Beta camcorder (via composite) be usefull?
Is yours analog? If yes, could you provide some more info about this camcorder (type, etc.)?

[Ookami]

*edit: I asume that you're talking about additions to BaronVlad's guide (or Wilbert's part in it etc.) if not, just forget my whole posting :) *

I don't want to lessen your fantastic contribution, but I just want to state that trading/sharing capture clips has nothing to do with serious testing. It can be a very small guiding point, but nothing to compare anything to.

Example.

I have an SVHS VCR, now I have a brand new retail VHS tape to backup it to SVCD (using a optimised TMPGEnc Pro). I use several filters etc. to make the backup as good as I can.

Person y has a crappy mono VHS VCR in the LQ area (playback only etc.) uses a 5th generation VHS wich almost cannot be capture without TBC and then uses a LQ program to encode to SVCD.

Both person used the same capture card... The difference should be quite visible ;) .

I think that every testing should be done under the same circumstance and done by the same user. Sharing of uncompressed clips is too unreliable (too much different factors different quality of drivers, OS, capture software, cables, playback devices (VCR etc.) ...) not to mention compressed clips where a few thousand factors come in additionally (used software and settings).

Thanks for listening :) .

Erm, just my 2000 cents.

Cheers,

Mijo.

[Wilbert]

*edit: I asume that you're talking about additions to BaronVlad's guide (or Wilbert's part in it etc.) if not, just forget my whole posting *
Only in the sense that we want to find out which capture cards/drivers perform scaling (NTSC) during capture, when and how.


I don't want to lessen your fantastic contribution, but I just want to state that trading/sharing capture clips has nothing to do with serious testing. It can be a very small guiding point, but nothing to compare anything to.
That's very true. I had no plans to do this ...


I think that every testing should be done under the same circumstance and done by the same user. Sharing of uncompressed clips is too unreliable (too much different factors different quality of drivers, OS, capture software, cables, playback devices (VCR etc.) ...) not to mention compressed clips where a few thousand factors come in additionally (used software and settings).
Actually, I was thinking about something else. But I won't tell you before BaronVlad put the new version up (and opens a new thread ...) :D

[FredThompson]

If standardized test discs are used with DVD players you could start to accumulate analysis of how the "default" settings capture color IF there is an automated sampling routine. Other than that, I'm not sure what can be done.

(I sell printing equipment. Everyone gets excited about printing a nice photograph but that's not a real test of the equipment. Same thing here...)

I still think a calibration method using some automated calculation would be a nice addition. You know, standardized frame in a loop, sample it 20 times, average them, see how far off "white" is, make a suggestion of what to adjust, that kind of thing. Sort of a way to test the "color" of the setup...

[Ookami]

@Wilbert

Aah, I thought you people got OT in the last few postings, so you're still OT ;) .

>Actually, I was thinking about something else. But I won't tell you before BaronVlad put the new version up (and opens a new thread ...)

LOL Well, I could get the infos from the boss, you know :D . Thanks for your great work! BTW, why is still the mega old version of the February on Doom9? Why isn't there a upload? Or should I empty my browsers cache ;) ?

@Fred

Nice idea. But... :D

Cheers,

Mijo.

[killingspree]

Originally posted by Ookami

BTW, why is still the mega old version of the February on Doom9? Why isn't there a upload? Or should I empty my browsers cache ;) ?


no it's not up yet... we've decided to not upload every little change seperately but do all the planned changes and upload it when it's ready. but in fact we are almost done anyway. i'd expect to see it up on doom9 very soon. as our 'boss' is rather busy atm, it might still take a week+ but if you talk to him and he ok's it i can send you a copy of the new and highly improved guide!

steVe

[Wilbert]

@jggimi,


I captured at 720x480 (1.5:1) YUY2, in HuffYUV with 172k PCM, and captured with an AverTV BT878 card, using the most recent - v5.3.6-1 - BTWinCap drivers with AVI_IO.

I'll make you unresized/unfiltered XviDs and e-mail them to you at the above address.
A while ago you sent those circle examples (720x480 and 640x480). Could you also mail them for 704x480?

@FredThompson,

I'm still patiently waiting for your samples :)

[jggimi]

If I recall correctly, I couldn't do 704x. I'll check again and let you know.

[FredThompson]

Yes, thanks for the reminder. I need to get samples done anyhow for another discussion thread. There's some web space on Geocities now and I'll put stuff there. I'll get on it.

[FredThompson]

I should have your test samples done today. There are two video segments I've been searchign for but haven't been abel to find. One was a Caucasian guy in white walking up a sand dune, the other was a show about Caesar's bridge over the Rhine (I think it was the Rhine) which had a lot of good sun on moving water. Have gone through about 20 tapes trying to find that source and the tapes must be gone. All the hardware is hooked up and I've now got a VHS copy of a Ken burns documentary about the Lewis and Clark exhibition. Sinch they travelled waterways, there should be some good footage there. The All-in-Wonder has never really done a good job with the really bright/pale stuff in NTSC. The card is supposed to handle PAL as well but I just got a PAL VCR so couldn't have tested that if I wanted to before. The only PAL source I have is camcorder footage from inside manufacturing plants. It's not a good sample.

So, I should be able to send you decent grabs int he next 24 hours or so. Gonna pull the Lewis & Clark thing 3 ways to compare, All-in-Wonder, DV passthru, and external MPEG2 encoder.

[SeeMoreDigital]

Fred

I've got some a high bit (9000kbps) Mpeg2 video of a Jaguar promo I can mail to you.

It's in both 4:3 and 16:9 aspect ratios, lasting 4min 34secs. It's great for testing because there a good deal of colour changes, colour burst, white burst, scene movement and some CGI.

Let me know if you want a copy as it fits onto a regular CD-R.

[FredThompson]

sure, check you private messages for a mailing address.

[SeeMoreDigital]

Job done and mailed!