Video Capture Quality Comparison (Chipsets)

[ozymango]

Having the original source is obviously a plus, but we can judge the quality of different captures anyhow, based of the main paramters

Well sure we can spend a lot of time (and have a lot of fun!) going back and forth on which video capture cards *we* prefer, but kinda trying to stay somewhat close to the OP's original question: "What capture card to use based on certain chipsets," and based on the chipsets he's got to work with, I'd go for the SAA713X. Again, limiting ourselves to what his real-world capture options are. I believe later on he says he has to work with composite video at some point, can't use Svideo for some reason? That's a big concern right there, might make a large difference to his end results.

Again I'm not against having fun running all sorts of tests on various cards, but really the only way to even compare those kinds of results (viewing test patterns) means capturing not just snapshots, but actual video footage, and then saving that footage at whatever compression rate you're wanting to test, and then uploading these samples to some third-party site so we can then download them to our own computers and watch the original files in their original form and ... well, again, I ain't opposed to that for entertainment purposes, but if the goal is to ever actually get around to digitizing precious videotapes for actual viewing for an audience -- might get around to that in 2027.

[aramkolt]

This is somewhat subjective since I'm not nearly as knowledgeable as the others here, but one thing a lot of cards struggle with is chroma bandwidth (those three sets of alternating vertical color bars in the bottom center) - To my eye the CX32885 has the best chroma bandwidth on a good signal (coming right off of the DVD player) and seems to have some of the fewest artifacts in the various diagonal line patterns. Putting it through a TBC first kind of a lot of the usefulness of the of the test to see how well the cards handle signals on their own since the output is largely based on what the Panasonic device has done with the signal first. The Phillips obviously has the worst chroma bandwidth. Since VHS isn't great to begin with when it comes to chroma, you don't want to be throwing away the little information you do have in my opinion, could be somewhat down to chroma saturation, but the levels just seem pretty off with the Phillips in particular.

Even looking at your original test captures, the skin tones on Jamie from the Mythbusters clip appears to be the most natural with expected level of sharpness with the CX card and I'd say it also has sharper color transitions/sharpness when you look at something like the tartan bars pattern or how well defined the large block of red is in the lower right of the SW2 pattern coming off of the DVD player. If it can't do well with a stable signal, it's not going to do well with an unstable one.

Short video clips are nice because it's easy to look at vectorscope/waveform monitoring to see if the captured chroma/luma levels are where they should be. You can also see chroma noise very easily in a vectorscope which isn't super apparent on still frames.

[ozymango]

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

You've done a heck of a lot of work to compare capture cards, to which I salute you! And as long as you're doing all that work, would you consider also testing the same captures using Virtualdub instead of DScaler? I've always used Virtualdub for my VHS captures but I'll also try DScaler to see if there's any noticeable difference in my capture results. This is not to disparage DScaler, but again for the sake of thoroughness, I'm curious of the different software would affect capturing quality results.

[Sharc]

@V2000: Did you have the Comb Filter of the passthrough device (DMR EH575) enabled in your VHS-Composite tests? The (2D/3D) comb filter should mitigate the dotcrawl/rainbow artefacts without blurring the picture.
Also, DScaler seems to have a Comb Filter. I have never tried it though ....
https://deinterlace.sourceforge.net/Help/TemporalComb.htm

(P.S.: I added a footnote to post#30 concerning the BT878 chipset.)

[Sharc]

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.

Sure, "bandwidth-limited" indirectly by specifying rise-/falltimes and shapes (pulse shaping filters) of the slopes/edges. Essential for unified and reproducible testing and development of test equipment. Makes sense, no doubt. My pragmatic point was simply that it is not forbidden to subject devices to other conditions and see how they deal with it. Therefore I wouldn't just disqualify the Belle Nuit test picture.
One can study test reports of shock absorbers which were subject to standard test procedures emulating certain (standardized) road conditions, and one can mount these on one's car and drive on a rough road of choice, and draw your own conclusions....

[lollo]

I'm curious of the different software would affect capturing quality results.

Unless they screw up the video architecture (i.e. OBS, software deinterlacing on the fly, no respect of the specs), the capture softwares have no impact on the (intrinsic) quality of the capture.

Captures done with AmarecTV, VirtualDub, VirtualVCR, my own capture software are identical (within the limit of the repeatability of an analog input signal across mutiple readings)

[V2000]

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

You've done a heck of a lot of work to compare capture cards, to which I salute you! And as long as you're doing all that work, would you consider also testing the same captures using Virtualdub instead of DScaler? I've always used Virtualdub for my VHS captures but I'll also try DScaler to see if there's any noticeable difference in my capture results. This is not to disparage DScaler, but again for the sake of thoroughness, I'm curious of the different software would affect capturing quality results.

I can't get my BT878 to work in VirtualDub because it has a wierd driver. The card i'm using was originally designed to be used as a frame grabber for cameras in the medical field. And Dscaler works really well with this card. I can even turn off the Chroma AGC and Comb Filter, which i did in my tests

[V2000]

@V2000: Did you have the Comb Filter of the passthrough device (DMR EH575) enabled in your VHS-Composite tests? The (2D/3D) comb filter should mitigate the dotcrawl/rainbow artefacts without blurring the picture.
Also, DScaler seems to have a Comb Filter. I have never tried it though ....
https://deinterlace.sourceforge.net/Help/TemporalComb.htm

(P.S.: I added a footnote to post#30 concerning the BT878 chipset.)

Hello! i had the NDR-Filter turned off on the Panasonic DMR EH 575. I remember it blurring the image a little too much for my liking on my DMR EH 52.
Besides that, wouln‘t be it a better idea to capture the image as authentic as possible so you can denoise using modern software?

[V2000]

@V2000: Did you have the Comb Filter of the passthrough device (DMR EH575) enabled in your VHS-Composite tests? The (2D/3D) comb filter should mitigate the dotcrawl/rainbow artefacts without blurring the picture.
Also, DScaler seems to have a Comb Filter. I have never tried it though ....
https://deinterlace.sourceforge.net/Help/TemporalComb.htm

(P.S.: I added a footnote to post#30 concerning the BT878 chipset.)

Hello! i had the NDR-Filter turned off on the Panasonic DMR EH 575. I remember it blurring the image a little too much for my liking on my DMR EH 52.
Besides that, wouln‘t be it a better idea to capture the image as authentic as possible so you can denoise using modern software?

Also i forgot to mention - DScaler does have some good Filters but as far is i know, none of them are applied to the recording itself. The they only filter the output to the screen

[Sharc]

@V2000: Did you have the Comb Filter of the passthrough device (DMR EH575) enabled in your VHS-Composite tests? The (2D/3D) comb filter should mitigate the dotcrawl/rainbow artefacts without blurring the picture.
Also, DScaler seems to have a Comb Filter. I have never tried it though ....
https://deinterlace.sourceforge.net/Help/TemporalComb.htm

(P.S.: I added a footnote to post#30 concerning the BT878 chipset.)

Hello! i had the NDR-Filter turned off on the Panasonic DMR EH 575. I remember it blurring the image a little too much for my liking on my DMR EH 52.
Besides that, wouln‘t be it a better idea to capture the image as authentic as possible so you can denoise using modern software?

I am talking about the Comb Filter for separating luma from chroma, not the noise reduction filter. Your Composite signal needs a good separation between luma and Chroma (Y/C) to minimize rainbows and dotcrawl which are strongly present in your composite captures. The chips/cards which you are testing appear to have poor Y/C separating filters only, so you have to switch in an external Y/C separating filter. DVD recorders usually have reasonably good Comb Filters for that purpose. So using the EH575 in passthrough you should just switch its Comb Filter ON.
I don't have the EH575 but I think it has this Comb Filter functionality available. Maybe it's called 3D Comb Filter, sharpness filter or similar, or some dull language dependent translation, I don't know for your model. Consult the manual. In any case a good external Y/C separation filter is usually required for these chips, before the composite signal enters the chip.
Signal flow is then VCR player Composite OUT -> EH575 Composite IN -> EH575 S-Video OUT -> Capture card S-Video IN.
Maybe someone with an EH575 may help with its menu settings.

The benefit of S-Video (which is preferred anyway) is that it doesn't require this Y/C filter because luma and chroma are conveyed on individual wires (Y and C).

[rgr]

I don't have the EH575 but I think it has this Comb Filter functionality available.

Yes, it has. Comb Filter option.

[ozymango]

I can't get my BT878 to work in VirtualDub because it has a wierd driver. The card i'm using was originally designed to be used as a frame grabber for cameras in the medical field. And Dscaler works really well with this card. I can even turn off the Chroma AGC and Comb Filter, which i did in my tests

Thanks for that update, I was just curious as to why you chose DScaler over Vdub. And yeah my first video "capture" card was a no-name card with a BT878 chipset that I'm pretty sure was used as a frame-grabber for medical cameras -- I only ever got it to work at 352x240 or whatever the lowest acceptable Video-CD rate was/is, and that was back when my computer monitor itself only went to 800x600.

[pandy]

Also i forgot to mention - DScaler does have some good Filters but as far is i know, none of them are applied to the recording itself. The they only filter the output to the screen

You may consider to use different Bt878 drivers (i recall some open source driver), i would try also to disable "peaking" filter as seem Bt/Cx samples provided by you looks like oversharpened, on other side i see that SAA is overcompressed - not sure why.

Bt and Cx may use different clock sources but yes, usually Bt may offer higher sampling rate than 13.5MHz (ITU) thus it may be reasonable to set square pixel parameters (like 768 pixels in line), Bt/Cx offer also other possibility - RAW grayscale sampling and eventual software decoding.