Video Capture Quality Comparison (Chipsets)

[V2000]

Happy new year everyone!

I'm trying to digitize a few of my VHS/Beta tapes and i was wondering which method would give me the best picture quality.

This is my Setup:

For tape playback: Panasonic NV-FJ730 and Sony SL-HF950.

They are hooked up to my Panasonic DMR EH 575 via Scart (carries Composite Video), which i use as a TBC.

The DVD Recorder is then connected to the following capture devices via S-Video:

1. Medion CTX 953 (Creatix SAA7131 Chipset)
2. IDS FALCON/EAGLE Frame Grabber (BT878 Chipset)
3. Hauppauge ImpactVCB-e (Conexant CX23885 Chipset)
4. Cheap eBay HDMI Capture Card (Connected to the HDMI Output)

I captured the videos using DScaler, encoded in Lagarith Lossless Codec.

According to the attached screenshots - which capture card would give me the cleanest picture?
Are there any modern, cheap alternatives?
I plan on deinterlacing with QTGMC.

I would greatly appreciate your opinion on this.



[Attachment 90607 - Click to enlarge]
Scene 1 - BT878


[Attachment 90608 - Click to enlarge]
Scene 1 - CX23885


[Attachment 90609 - Click to enlarge]
Scene 1 - HDMI


[Attachment 90610 - Click to enlarge]
Scene 1 - SAA 7131





[Attachment 90611 - Click to enlarge]
Scene 2 - BT878


[Attachment 90612 - Click to enlarge]
Scene 2 - CX23885


[Attachment 90616 - Click to enlarge]
Scene 2 - HDMI


[Attachment 90614 - Click to enlarge]
Scene 2 - SAA7131





[Attachment 90615 - Click to enlarge]
Scene 3 - BT878




[Attachment 90617 - Click to enlarge]
Scene 3 - CX23885


[Attachment 90618 - Click to enlarge]
Scene 3 - HDMI


[Attachment 90619 - Click to enlarge]
Scene 3 - SAA7131

[aramkolt]

I applaud your efforts - I've been meaning to compare like 15 different capture cards or methods against each other, but I stumble on the actual test to perform.

My suggestion would be to burn this DVD image (it's very small, something like 65MB and it loops) - it's a lopping short segment of the SW2 pattern with moving zone plate by Snell and Wilcox.

From there, use each card to capture directly from a DVD player that is playing it and post screenshots from those captures. Ideally you'd capture via composite and then S-Video to see how well their comb filters work in each card.

From there, I'd make a VHS recording via S-Video off of the DVD player of that same pattern and then capture from the VHS tape via S-Video for each card.

Doing it this way will identify all kinds of video problems much better than random screencaps as this particular pattern is designed to reveal video reproduction weaknesses.

[V2000]

My suggestion would be to burn this DVD image (it's very small, something like 65MB and it loops) - it's a lopping short segment of the SW2 pattern with moving zone plate by Snell and Wilcox.

Hello aramkolt, i've noticed that the DVD image contains a NTSC recording. Are you able to provide me with a suitable PAL iso?

Thanks.

[ozymango]

Happy New Year to you! And everyone else also!

Looks like you're having fun with your capture tests, and all I can offer you is this -- I'm working on digitizing a bunch of VHS tapes myself, all home recordings (e.g home VHS camcorder recordings, and even some old video camera and portable VHS VCR recordings), I have a JVC HR-S4800U SVS VCR for playback, I pass the Svideo output into a Panasonic DMR-ES25 dvd recorder (works great as a TBC), and then I import the output using VirtualDub using a Compro Videomate PCI capture card (Philips SAA7133 chip), HuffYUV codec ... oh, yeah, it's a Dell Vostro 4-core Pentium computer running Windows 10 Pro LTSC ... then I run the avi files through Kdenlive to make videos I can watch on my TV -- I don't deinterlace anything, I make these videos for my TV playback -- and you know something, it works pretty well. I'm satisfied with the quality, anyway.

For the record I also have a cheap Intel capture card I can't quite ID, that's a PCI card with a BT878 chipset, I have a Hauppage WinTV HDR-1250 PCIx card (Conexant CX23885 chipset), an ATI TV Wonder 600 PCI card with (I think) the ATI Theater chipset, and a cheap HDMI-to-USB capture device, and I've tried capturing VHS videos using all those devices, I've used VirtualDub, I've used some software that I can't remember that works with the ATI card, and I've used OBS to capture pretty much every device, and in the end -- I like my Compro Videomate setup, it looks maybe just a bit sharper than any of the other methods -- but frankly any of them work in a pinch. But it's fun to play around and see what I like best.

Good luck!

Edit: One thing I've encountered that makes it impossible to offer The Answer to the question of "What looks best" is -- it just depends on the original tape, and the original VCR, and maybe sunspots, and some tapes looks better than others, some tapes were recorded while the camera was being tossed in the air and every shot makes you seasick, some tapes look great and you can't hear anything anybody says, and some tapes have great audio but everybody is backlit so all you can see are silhouettes ... anyway, again good luck and one thing is, if these are vintage tapes with a cast of friends and family who've passed on, any way you can watch them again, just for the memories, is worth it.

[aramkolt]

Ah, I hadn't picked up that you are in PAL land. That source MPEG file comes from the digital video essentials DVD which is the NTSC version, but I haven't seen it posted for the PAL version of the DVE disc.

I do have a real TPG21 that makes this pattern (and it can do both PAL and NTSC), but there's really no way to "digitally" get the pattern off of it. It'll only output SW2 pattern via either composite or S-Video oddly enough.

I'll see about putting it in PAL mode and then capturing the pattern with a Brighteye BE3 which should result in a pretty accurate interlaced capture via ProRes422HQ I would think.

I'll post back when that's completed with a PAL type ISO

[aramkolt]

Here's the PAL pattern DVD I made with the test pattern generator. I believe the way the disc is authored that it should just loop continuously, or you could just set your DVD player to "Loop" a chapter if you want a certain pattern to stay on longer. When you're making a test VHS tape out of it from the DVD player, you can just let the DVD player keep playing. It is important to record the "test VHS tape" using S-Video output of the DVD player or you'll get a bunch of rainbowing on all tests done with the VHS tape.

I suspect the differences will be much more obvious running these patterns through your cards.

My recommendations for testing cards are as follows:

1. Burn the ISO onto a recordable DVD
2. Capture directly from the DVD player's S-Video output AND a separate composite recording for each card
3. Make a VHS "test tape" by recording the patterns via S-Video off of the DVD player output.
4. Capture the "test tape" via S-Video with each card

The reason you don't really need to try composite again for the "test tape" is that you'll already know the "ideal" comb filter performance from the direct from DVD player composite capture done in step 2.

Also attached an explanation of the SW2 pattern areas and what they are supposed to look like.

My guess is you'll be pretty shocked at how poor chroma frequency response is on VHS and just how blurred the different fine lines will be.

[Sharc]

FWIW here a SBS comparison between Live2 and GV-USB2.
Live2 procamp settings: default. Note the white at Y~230/RGB~250
GV-USB procamp settings: contrast reduced from default 105 to 100 in order to set the white at Y~235/RGB~254.

Edit: To keep the uploadable filesize <500MB I compressed the comparison using x264 crf 10, so some minor compression artefacts survived ....

[aramkolt]

Sharc - since you know waaaaay more about analog video than I do, are there any other specific patterns that you think would be useful when comparing capture card performance that you'd expect some cards to struggle with?

I do plan to throw in an audio sync test also to give an idea of how much video delay if any a chain produces.

One "defect" I've seen before in some capture chains is where the chroma is either delayed by a frame or possibly gets "averaged" due to some type of noise reduction or 3D comb filtering perhaps. It seems that luma/chroma/or both get averaged across frames, so you'll get kind of a "chroma ghost/residual chroma ghost image" right at the first frame of a sharp cut-scene transition - but I'm not really sure which pattern would test for this. The TPG21 can alternate between two static patterns as fast as you want, so perhaps going from say color bars to plain white back and forth might show this potential defect/side effect of processing?

The other thing I really wanted to test on all cards is what the true chroma subsampling and chroma bit depth is in any given chain. I think a shallow ramp will show "banding" when 8 bit vs 10 bit is used for capture, but still not sure which pattern would show whether say 4:2:0 subsampling is happening somewhere in the chain, particularly if the medium is a DVD which is 4:2:0 and 8 bit to begin with.

I have ways of getting a 10 bit 4:4:4 stream from my computer into analog, but it's definitely not going to be as convenient as a burnable DVD for others to try on their chains.

Would love to hear your thoughts!

[jagabo]

The Belle Nuite chart is another good test:
https://www.belle-nuit.com/test-chart

[aramkolt]

Thanks for the suggestion! I agree, that that the Belle Nuit pattern will probably show some defects better than SW2 alone.

I wasn't exactly clear on how to turn the pattern Tiff into an MPEG2 file, so I put it into Davinci Resolve on a PAL type timeline, exported as ProRes422HQ and then put it into my MPEG encoder. I'll attach both the ProRes file and the resulting MPEG - you can see that the fine lines definitely aren't as clear.



I still don't know if there's a good way to identify 4:2:2 vs 4:1:1 vs 4:2:0 chroma subsampling in the chain, so would love to know of any ideas for that. The method I planned to use for myself involves a Davinci timeline and then going through a Blackmagic Monitor 3G (which outputs SDI) and then an SDI to analog converter which can output composite, component, or S-Video. However, I want a test that is more universally accessible without users having to invest in that sort of hardware to test their chain. Guessing using DVD as the medium to test probably is not possible though since that's inherently 4:2:0 and 8 bit. So the question is, is there an easy reproducible way to get 10bit/4:2:2 playback of such a file that can be fed out as composite/Svideo from a relatively ubiquitous device?

[lollo]

and in the end -- I like my Compro Videomate setup, it looks maybe just a bit sharper than any of the other methods

Difficult to believe that a ATI USB 600 capture is lower quality than a Compro Videomate. Do you have some sample to post?

Concerning the general topic, in addition to clean input DVD signals, do not forget also to compare the cards in the real use-case of VHS/S-VHS capture (with a clean and time base corrected signal), which is much more important

[jagabo]

The Belle Nuit charts are designed to be converted to YUV with a pc.601 matrix -- ie, without the usual rec.601 levels compression that is normally used for that conversion. That gives you proper levels in the YUV domain, black Y=16, white Y=235. Upon conversion back to RGB you would use the usual rec.601 conversion with contrast stretch. That will crush the superblacks and superwhites (ie, Y=0, 4, and 12 will all be R=G=B=0, and Y=255, 251, and 239 will all be R=G=B=255), and the main color patches should have RGB primaries of 191 or 0. Rounding errors at each of those steps will give slight errors in those values.

AviSynth script to load Belle Nuit RGB and convert to standard rec.601 YUV:

ImageSource("testchartpal.tif", start=0, end=124, fps=25) 
ConvertToYUY2(matrix="pc.601")

resulting uncompressed rec.601 YUY2 AVI attached in a ZIP file.

[ozymango]

Difficult to believe that a ATI USB 600 capture is lower quality than a Compro Videomate. Do you have some sample to post?

I didn't say ATI 600 USB, I said ATI TV Wonder 600 PCI. Ask lordsmurf what he thinks of the ATI TV Wonder 600 PCI for capturing.

The Compro Videomate uses the Philips SAA7130 chip for AV processing, and that chip is quite well reviewed. You don't like it, use a different capture device.

As for posts of sample video, I'd do it except two things: 1) I'm lazy , and 2) unless you could actually watch the original VHS tape as played back from my SVHS deck onto a TV, you wouldn't really be able to say if the capture is decent or not because you won't have any clue what the source material looks like. And even when I capture as close to the "original" as possible, I then go through it afterwards and tweak thing to my satisfaction. So it won't do you any good to know what my samples look like, you're just gonna have to get your own equipment and see if it works to your satisfaction or not.

[aramkolt]

That's the beauty of patterns! We do know how they should look and what the different chroma/luma levels should be. The purpose of capturing off of the DVD player first is to see what capture devices do with clean signals (no timebase errors) and before any significant quality loss from the bandwidth limitations of the analog formats themselves (which you will see is quite significant when comparing a VHS test tape of the same pattern).

For other issues like dropped frames and unstable signals (which aren't present on DVD player output), patterns can't help much unfortunately. I do think there's a use for a visibly burnt in timecode on each frame possibly combined with the equivalent timecode audio. Doing that actually will tell you how far ahead the audio is of the video and would allow you identify duplicated or dropped frames after a capture has already been done. Others here may have a better way of identifying duplicated frames, but dropped frames may be more difficult to identify for software depending on the capture setup.

If anyone has ideas for how to generate time code audio and a visible timecode overlay on each frame of something, I'd like to know about it. The visible timecode part I don't think is difficult and could be done in Resolve or similar.

I actually hadn't seen this tool, but it's apparently a free one that will make timecode audio at the duration of your choice:
https://elteesee.pehrhovey.net

Careful - actually listening to the timecode audio will make your ears bleed

From there, I think you'd just add the audio to your audio sync test DVD timeline, have the nonlinear editor interpret it into a visible timecode, then export that file to be put on DVD. Then simply record the DVD with a VHS player and capture the VHS to see if the audio drifts over time or not.

Would like to know if anyone sees a flaw with that or if there's some type of lack of usefulness in doing that to identify dropped frames after a capture has already been made of it. The idea is that at the beginning of the capture, the visible timecode and timecode audio should have a certain gap that remains fixed throughout the capture if there's no dropped frames. If there are dropped frames that aren't followed by a null or duplicated frame, the gap between the displayed timecode and the audio timecode. This could also provide useful information on exactly how much audio delay should be added to captures of that same chain to get the audio to exactly sync to the video.

[rgr]

According to the attached screenshots - which capture card would give me the cleanest picture?

This is a poor choice. Cleanest doesn't necessarily mean best. First, check which method doesn't clip blacks and whites (histogram).

[jagabo]

If making your own test videos you can add timecodes with AviSynth.

ImageSource("testchartpal.tif", start=0, end=124, fps=25) 
ConvertToYUY2(matrix="pc.601")
ShowFrameNumber(x=106, y=32)
ShowTime(x=156, y=555)

[Attachment 90657 - Click to enlarge]

You can see the frame number (50 in the image) near the top left and the frame time (00:00:02.000) near the bottom left.

[Sharc]

....I still don't know if there's a good way to identify 4:2:2 vs 4:1:1 vs 4:2:0 chroma subsampling in the chain, so would love to know of any ideas for that....

Just based on the final result it is usually not possible to conclude on subsampling steps which were made somewhere along the processing chain. One can however use avisynth's UtoY() and VtoY() functions to visualize the actual chroma subsampling, and also to discover flaws like incorrect interlaced 4:2:2 to interlaced 4:2:0 conversions, for example.

[aramkolt]

I was able to figure out how to make the frame numbers and timecode visible in Davinci Resolve, but I wasn't able to figure out how to get Resolve to interpret a separately added audio track as LTC audio.

The idea would be to add the LTC audio first, then reassign the test video's timecode to be that of the timecode audio, then display the timecode. In this way, the timecode audio should exactly match the displayed timecode and then when capturing, you could compare the lag in the timecode audio vs the displayed timecode.

I guess technically one could just have a cyclical beep every minute and then later see if the beeps ever start drifting from the realtime capture clock, it just doesn't seem as precise.

[jagabo]

The above YUV Belle Nuit chart with the Y, U and V channels displayed as 8 bit RGB, using:

StackHorizontal(GreyScale(), UtoY(), VtoY())
ConvertToRGB(matrix="pc.601")

[Attachment 90658 - Click to enlarge]

Y on the left, U, in the middle, V on the right. Note, the Y channel is 720x576, the U and V channels are 360x576. YV12 video would have a 720x576 Y channel and 360x288 U and V channels.

[Alwyn]

FWIW here a SBS comparison between Live2 and GV-USB2.
Live2 procamp settings: default. Note the white at Y~230/RGB~250
GV-USB procamp settings: contrast reduced from default 105 to 100 in order to set the white at Y~235/RGB~254.

@Sharc, Is your GV Contrast reduction to 100 a general "rule" for you or only applicable to that test?

[Sharc]

FWIW here a SBS comparison between Live2 and GV-USB2.
Live2 procamp settings: default. Note the white at Y~230/RGB~250
GV-USB procamp settings: contrast reduced from default 105 to 100 in order to set the white at Y~235/RGB~254.

@Sharc, Is your GV Contrast reduction to 100 a general "rule" for you or only applicable to that test?

Yes, almost a "general rule" here for the GV-USB2 to put the "reference white" of testpatterns at Y=235 -> RGB(255,255,255) and the colors of standard (SMPTE) colorbars close at their "191 or 0" nominal RGB on the monitor when using the standard (expanding) Rec.601 decoding matrix.
It is usually a good setting for real tape captures as well to prevent the brights from crawling too much into the luma 236....255 range (except halos and other artefacts).

[Alwyn]

Thanks Sharc, I understand the "generality".

[pandy]

The Belle Nuite chart is another good test:
https://www.belle-nuit.com/test-chart

Nope - this is very bad test pattern - signals present in this pattern are not bandlimited and as such they can introduce problems not present in real signals.

[lollo]

I didn't say ATI 600 USB, I said ATI TV Wonder 600 PCI. Ask lordsmurf what he thinks of the ATI TV Wonder 600 PCI for capturing.

My bad. In any case, lordsmurf and I have very different judgement about the capture cards, so he's not a reference for me

The Compro Videomate uses the Philips SAA7130 chip for AV processing, and that chip is quite well reviewed. You don't like it, use a different capture device.

Of course!

you wouldn't really be able to say if the capture is decent or not because you won't have any clue what the source material looks like.

False. Having the original source is obviously a plus, but we can judge the quality of different captures anyhow, based of the main paramters (I won't mention any, apparently you are not sensitive to that).
Nothing really important anyhow, it was just a curiosity of mine.

[lollo]

It is usually a good setting for real tape captures as well to prevent the brights from crawling too much into the luma 236....255 range (except halos and other artefacts).

For many of my tapes is a good setting to avoid clipping whites to Y>255, not just to stay in Y limited range! GV-USB2 has a natural tendency to capture in the high brigthness level.

[Sharc]

The Belle Nuite chart is another good test:
https://www.belle-nuit.com/test-chart

Nope - this is very bad test pattern - signals present in this pattern are not bandlimited and as such they can introduce problems not present in real signals.

Well yes, but it always depends on what we want to test. One may intentionally want to subject capture cards to some "torture" test and see how they cope with "unusual" (out of regular range in some respect) signals, even though one might assume that such signals should not be present in real systems.
(The correct interpretation of the tests outcome is a subject of its own).

[Sharc]

It is usually a good setting for real tape captures as well to prevent the brights from crawling too much into the luma 236....255 range (except halos and other artefacts).

For many of my tapes is a good setting to avoid clipping whites to Y>255, not just to stay in Y limited range! GV-USB2 has a natural tendency to capture in the high brigthness level.

Your milage may vary depending on the actual situation. That's why I used terms like "usually" and "generally" and the well defined "reference white" of standard test pictures.
More caveats exist for other capture cards, especially for those doing some unexpected "auto" adjustments and causing unnecessary high C-wire -> Y-wire crosstalk due to their insane 'high-impedance' C-wire termination, almost doubling the chroma amplitude in the S-video cable compared to nominal 75 Ohms termination .
One can see the "ants" for the Live2 in the solid colors of post#7 attachment.

[pandy]

Well yes, but it always depends on what we want to test. One may intentionally want to subject capture cards to some "torture" test and see how they cope with "unusual" (out of regular range in some respect) signals, even though one might assume that such signals should not be present in real systems.
(The correct interpretation of the tests outcome is a subject of its own).

If there is some compression (using DCT or similar transformation) then such stripes is asking for trouble - doubt if there is way to distinguish between source signal and converter problems - Snell Wilcox test is bandlimited not by accident also any sane test signal standard require bandlimited stimulus - even simple color bar pattern has particular requirements on this - just check ITU-R BT.801 for digital encoding test signals - bandwidth defined and limited.

[V2000]

Hello! Thank you all for your input on this. I have done my homework and made some captures:


For the source i used a high end set-top-box with an integrated DVD Player (Dreambox 8000), this should give us a high quality, authentic output.


Direct S-Video Output:

Pattern 1:




BT878

[Attachment 90669 - Click to enlarge]



SAA7131

[Attachment 90671 - Click to enlarge]


CX23885

[Attachment 90672 - Click to enlarge]


Pattern 2:


BT878

[Attachment 90673 - Click to enlarge]


SAA7131:

[Attachment 90674 - Click to enlarge]


CX23885:

[Attachment 90675 - Click to enlarge]


Pattern 3:

BT878:


[Attachment 90676 - Click to enlarge]



SAA7131:

[Attachment 90677 - Click to enlarge]


CX23885

[Attachment 90678 - Click to enlarge]




But since i don't really have a way to capture VHS over S-Video (sold my only S-VHS Recorder years ago), i did a second direct capture using Composite:

Pattern 1:

BT878 -

[Attachment 90679 - Click to enlarge]

SAA7131:

[Attachment 90680 - Click to enlarge]

CX23885:

[Attachment 90681 - Click to enlarge]


Pattern 2:

BT878

[Attachment 90682 - Click to enlarge]


SAA7131

[Attachment 90683 - Click to enlarge]


CX23885:

[Attachment 90684 - Click to enlarge]


Pattern 3:

BT878

[Attachment 90685 - Click to enlarge]

SAA7131

[Attachment 90686 - Click to enlarge]

CX23885

[Attachment 90687 - Click to enlarge]



And now to the VHS Recording, captured using the Panasonic DMR EH 575 as a passthrough TBC:

Pattern 1:

BT878

[Attachment 90688 - Click to enlarge]


SAA7131

[Attachment 90689 - Click to enlarge]


CX23885

[Attachment 90690 - Click to enlarge]



Pattern 2:

BT878


[Attachment 90691 - Click to enlarge]


SAA7131

[Attachment 90692 - Click to enlarge]

CX23885

[Attachment 90693 - Click to enlarge]



Pattern 3:

BT878

[Attachment 90694 - Click to enlarge]


SAA7131

[Attachment 90695 - Click to enlarge]


CX23885

[Attachment 90702 - Click to enlarge]

I have noticed that the Panasonic DMR EH 575 does a great job at stabilizing the image but leaves a noticeable amount of sharpness behind.
The Picture looks noticeably less sharp even though the DNR feature is disabled.

So what Capture card would you use?


I have also gotten my hands on a Sony DCR TRV 345 Digital8 camcorder, which has an analog input that can be passed to the Firewire port.

Unfortunately i can't get it to work. When i connect the Firewire cable, it shows AV > DV on the Display so it recognizes that there is something plugged in but my PC will not recognize it. It will simply not show up in WinDV. I have tried it on Windows XP and Windows 7 32-Bit.
Am i missing something? I'm guessing the USB Streaming will give me a highly compressed image.

[Sharc]

So what Capture card would you use?

Definitely not the BT878.

Just briefly:
BT878: Colors are offset / wrong. Capture window (aspect ratio) looks incorrect (slightly stretched horizontally and cropped on top). *)
SAA7131 and CX32885: Better than the BT878 but not convincing. Maybe it is also your setup.

VHS Composite video: Typical composite artifacts, means rainbows and dotcrawls. A useful Y/C separating filter (2D/3D comb filter) seems to be missing or it does a poor job. Horizontal resolution is poor (even for VHS).

And screenshots don't tell too much. One would have to inspect short clips.

*) Footnote:
As far as I now the BT878 chipset did not sample the luma at the standard (Rec.601) 13.5MHz but something higher (14.75 MHz for square pixels?), means the Pixel Aspect Ratio does not comply with Rec.601. To compensate this, they scaled the picture in post somehow, resulting probably in the cropping of the top and slight horizontal stretching. Still not compliant with Rec.601 though, IMO.