New Technique: Component4All!!

[jmac698]

Yet another ground-breaking cap technique: now anyone can capture pure component video with no special equipment! Forget expensive pro-level cards, you don't need 'em!

Updated 03-2006:
This technique increases your potential resolution by about 2x beyond what you could do with lossless HuffyUV. So the real resolution is 1440x480, yes that's high-def! To keep this amazing resolution you will have to save in hi-def, otherwise, you will lose - for example saving as Huffyuv will use YUY2 color format - this means 2 pixels of color are stored as one.

Caveats:
- your source must be 4:4:4 ** UPDATE: this technique always improves quality, because svideo has inherently low color resolution, even if your source was only YUY2 orignally. You can get all the 4:4:4 quality as well if the source uses real chroma or 4:4:4 chroma upsampling.
- you will have to cap 3 times in a row, in a repeatable way
- it's slow
- the cheap and fast version, has slightly wrong colors
- you have to really want some hi-def glory...

Ok, now here's how: you need a "Y" splitter cable, two females to one male, type cinch/RCA/phono. You need an svideo to composite connector, or just a composite connector on your cap-card.

DISCLAIMER: This technique requires you to make non-standard connections. The risk is your own.

Connect Y directly to your composite in. Cap. Connect Y and Pb together, send to your composite in. Cap. Connect Y and Pr together. Cap.

Make 3 files: y.avs, ypb.avs, ypr.avs. Put your mpeg2source and trim commands in here, so that all files start playing from the same frame.

Final step: run this script.
y=avisource("C:\MyVideos\y.avs")
ypb=avisource("C:\MyVideos\ypb.avs")
ypr=avisource("C:\MyVideos\ypr.avs")
vyu=mergergb(ypr,y,ypb)
return(vyu)

That's it! Your done. The final file is in full resolution RGB. Resize and convert to YUY2 as needed.

Results:
-The grey range, and flesh colors are excellent. Some rare colors are shifted, but you cannot notice this. In theory it's "wrong", but it works.
-The color fringes around strong colors are gone! (something I find very annoying about s-video).
-If you dropped any frames in any caps, you will start to see some ugly color ghosting.
- UPDATE: I now have a script which exactly recreates the RGB picture, colorbar test patterns are accurate. However this requires some careful, one-time calibration. I'm working on the calibration proceedure.
- UPDATE: I have test results which prove >=1.6 color resolution as measured from a DVD Video Essentials test DVD pattern - chroma frequency sweep.

I have many more ideas brewing - stay tuned. (>2 times further resolution is possible...)

[tedkunich]

This technique increases your potential resolution by 4x (2x in horizontal and vertical). So the real resolution is 1440x960, yes that's high-def! To keep this amazing resolution you will have to save in hi-def, otherwise, you will lose - for example saving as MPEG2 will use YV12 color format - this means 4 pixels of color are stored as one.



Expain to me how capturing and layering the multiple color planes is going to give me "HiDef" 1440x960 out of my trusty AIW capture card which is only capable of 720x480??? All you are getting out of this is "possibly" a better color depth in the final capture.

T

[jmac698]

I was waiting for someone to say that :)
Yes, of course it's debatable, but let me explain my reasoning.
If your card caps in CCIR656 format, that's 4:2:0 or 4 pixels/luma, your resolution is 720x480 Y, 360x240 U/V. If we call this "720x480 resolution", then any increase must be spoken in different terms. My technique caps 720x480 Y, 720x480 U/V. You cannot store that resolution in a mpeg2 without loss. The only way to keep this true color resolution is with 4x larger mpg file.

Or if you prefer, call it true 720x480, because usual 720x480 was never true. But be sure to call DVD quality not true 720x480 either.

Besides this debate, is the reality. I've tried it, and it does look better to me, even a test DVD with frequency tests shows better U/V response. And it's a hidef source sent at 4:4:4 (apparently). That just shows you - it takes a hi-def mpg2, downsampled, to make true 4:4:4 color.

What if I said I could also cap 480P? Is that an increase either?

There are also theories of perceptual resolution. Usually the saying goes, we can't see colors in hires. So how come I can see fringes of colors in s-video? It's too low color rez vs luma rez.

[mod]

Interesting jmac698, could you put somewhere some compressed frame to show the difference?

[tedkunich]

I was waiting for someone to say that :)
Yes, of course it's debatable, but let me explain my reasoning.
If your card caps in CCIR656 format, that's 4:2:0 or 4 pixels/luma, your resolution is 720x480 Y, 360x240 U/V. If we call this "720x480 resolution", then any increase must be spoken in different terms. My technique caps 720x480 Y, 720x480 U/V. You cannot store that resolution in a mpeg2 without loss. The only way to keep this true color resolution is with 4x larger mpg file.

Or if you prefer, call it true 720x480, because usual 720x480 was never true. But be sure to call DVD quality not true 720x480 either.

Besides this debate, is the reality. I've tried it, and it does look better to me, even a test DVD with frequency tests shows better U/V response. And it's a hidef source sent at 4:4:4 (apparently). That just shows you - it takes a hi-def mpg2, downsampled, to make true 4:4:4 color.

What if I said I could also cap 480P? Is that an increase either?

There are also theories of perceptual resolution. Usually the saying goes, we can't see colors in hires. So how come I can see fringes of colors in s-video? It's too low color rez vs luma rez.


I don't know what card or the source you are using, but both my AIW and my CX2881 based cards can both capture in YUY2 (4:2:2) format. I have never noticed any color fringing as you describe (primarily LD and OTA caps) using either card - I NEVER cap using any MPEG codec... I capture to AVI using HuffYUV.

As to your capture problem, it could be one of two things (as it always is with analog capture) your source device or your capture device. Your source could be smearing the chroma information, i.e. poorly designed or defective electronics, or your card is garbage, i.e. poorly designed or defective electronics. :) It does not take much of a phase delay (common in poorly designed electronics) to introduce artifacts in the chroma vs luma overlay. Cables with too much cross talk/leakage can also cause similar problems.

I think it would behoove you to read up on luma and color as used in the current video standards (NTSC or PAL) and see the relationship between them - more is not always better :) Improved bit depth (4:4:4 vs 4:2:2 vs 4:2:0) only delivers greater color accuracy, not detail.

The idea of using multiple captures to improve the final results is not a new idea - search the Avisynth forum for a filter called TooT and its derivatives. They perform a pixel by pixel averaging accross three caps and toss any outliers producing signifiantly reduced random noise in the final capture - the do nothing to improve the resolution of the end result.

T

[jmac698]

capture in YUY2 (4:2:2) format
Ok! I therefore have an unusual card, or else I don't know how to use it. I have a Hauppage PVR, which has a hardware mpeg encoder. Perhaps they knew it was going to MPEG2, so they only enabled 4:2:0 capture. I looked at the filter graph, and there is a "CCIR656" output, which is also 4:2:0. So far, I can't find a way to cap to YUY2.

Regardless, my technique causes a 2x improvement in that case, using AVI formats and a YUY2 card.

I am aware of the multiple cap technique, I read the original post a long time ago. This is not the same at all, obviously, but I could make use of it to fix small off-by-one errors in Y between my caps.

S-video fringing: have to explain this better, so we both know what we're talking about. Here's two examples: the beginning of Blade Runner, with the red word "Replicant", certainly stesses color resolution. From DVD to TV in svideo, also from Hollywood Plus to TV, I see smearing of the red, as a gradient, fading towards the right, for about a few pixels. It also shows up in OTA caps. You have to know what to look for, it's generally not a big problem, just when two pure colors are contrasted together. Look yourself, for a blue object on a white background, and see if there's any blue that leaked into the white.

I looked at my technique in both U and V planes with VToY(). I can read lettering in my "C4All" technique, but I can't in the U/V of Svideo cap.

Svideo uses a color carrier of 4.43/3.57MHz. It can't possibly have higher frequency content, because it is amplitude/phase modulated. Luma has a 13.5MHz rate or so, if you do 720*1/53.3uS. So recording full bandwidth color from the luma input pin certainly makes sense.

mproved bit depth (4:4:4 vs 4:2:2 vs 4:2:0) only delivers greater color accuracy, not detail.
A Technical Introduction to Digital Video 1st Ed., Charles A. Poynton, P25:

4:4:4 Y'CbCr

Y'0Y'1
Y'2Y'3
CbCb
CbCb
CrCr
CrCr

4:2:2
__Y'0Y'1
__Y'2Y'3
Cb__
Cb__
Cr__
Cr__

4:4:4, 24bits/pixel. 4:2:2, 16bits/pixel. 4:2:0, 12bits/pixel. There is 8bits to store Y/Cb/Cr, but eg. 12 bytes to store the 4 pixels of 4:4:4, averaging 24 bits/pixel.
I think you may be confused when they say 4:2:2 has more bits/pixel than 4:2:0, yes that's true, but it's only because every other pixel of Cb/Cr is taken, and they average to 16bits/pixel. Really, each value is still stored in the same color depth, 8 bits of Y/Cb/Cr.

It also shows that the color comes from different pixels than the Y ones! So the color is inherently offset by 1 pixel left. That would explain the color shift problem, which is corrected by the vhs filter, or the option in dscaler.

If it were rgb 12 bit graphics, it would only have 4096 colors, 4 bits per value, and look obviously much worse :)

[tedkunich]

Svideo uses a color carrier of 4.43/3.57MHz. It can't possibly have higher frequency content, because it is amplitude/phase modulated. Luma has a 13.5MHz rate or so, if you do 720*1/53.3uS. So recording full bandwidth color from the luma input pin certainly makes sense.


Ummm... there is no way that NTSC SD video has a Luma B/W of 13.5MHz. more like 4-5MHz, ~6 in PAL land.

And BTW, what source would you have that is outputing in component and yet is not already in a digital format and why would you want to capture this through an analog pipe? If it is off of a PVR or a DVD, either grab the .ts file or rip it - will certainly be faster and better results.

T

[Wilbert]

@jmac698,

You might want to read
http://www.doom9.org/index.html?/capture/introduction.html
http://www.doom9.org/index.html?/capture/analogue_video.html

[jmac698]

there is no way that NTSC SD video has a Luma B/W of 13.5MHz.

From the capture guide:
This makes your final sample rate 704 samples / 52 µs = 13.54 MHz.

This is sample rate. According to the Nyquist theorem, you can record frequencies of half the sample rate, making the bandwidth 6.77MHz. Modern video DAC's have excellent filters and I expect to reach very near this value. If you read my post, I said rate, so while I was correct, I should have compared bandwidth MHz numbers together, I apologize for the mistake. In any case, the luma bandwidth is still greater, in one estimation (approx.), 6.77/3.57=1.88 times better.

either grab the .ts file or rip it this would be ideal, but not everyone has an HDTV/DVB card. Also, many satellite settop boxes don't even have firewire. The point, obviously, is to record from a hidef channel on a satellite box in better quality. I suspect the svideo signal itself is also partly to blame.

Anyhow, the technique is there for those who want to make use of it, as is the multiple cap technique to reduce noise. Judging from the forums, many of us are interested in ways of improving quality.

[tedkunich]

From the capture guide:
This makes your final sample rate 704 samples / 52 µs = 13.54 MHz.

This is sample rate. According to the Nyquist theorem, you can record frequencies of half the sample rate, making the bandwidth 6.77MHz. Modern video DAC's have excellent filters and I expect to reach very near this value. If you read my post, I said rate, so while I was correct, I should have compared bandwidth MHz numbers together, I apologize for the mistake. In any case, the luma bandwidth is still greater, in one estimation (approx.), 6.77/3.57=1.88 times better.


Not trying to bust your chops bud, just trying to point out discrepancies. Everyone loves to quote Nyquist as if that was a holy grail or something... sure you can sample at that frequency... at a piss poor accuracy! Nyquist only determines MAXIMUM frequency that you can sample and NOT get aliasing - it does not state anything about the accuracy of that signal, i.e. voltage, which happens to be very important in video as it determines the brightness, contrast and detail of video.

Also do not forget that all cards on the market have some intended and unintended filtering happening in the analog domain before it ever reaches the ADC (not a DAC BTW). Since MFG's typically use a simple, ie inexpensive, 3db or 6db filter prior to the ADC, the cut-off frequency will be significantly lower than Nyquist such that the rolled off frequencies will be reduced below the ADC's sensitivity @ Nyquist. I would be willing to bet that your card is limiting the analog to something less than 4.5MHz, so no matter what you do RE sampling, you are never going to get better than that.

Since it sounds like you have a PVR (I assume since you mention a three pass capture), why not hack the box or remove the HD to get access to the files?



Peace,

T

[rfmmars]

Results:
-The grey range, and flesh colors are excellent. Some rare colors are shifted, but you cannot notice this. In theory it's "wrong", but it works.
-The color fringes around strong colors are gone! (something I find very annoying about s-video).


Using a "Y" cable can cause a problem with color because of the difference sometimes between the impedance of the two outputs.

It would be best to use a one transistor emitter follower for each source before inputing to the capture card. Many times the chroma channel out is of high impedance compared to the ~ 50 ohm impedance of the "Y" out.

This does in some cases causes color smear or phase shift.

I am still interested in your idea.

Richard

[jmac698]

rfmmars:

Using a "Y" cable can cause a problem with color because of the difference sometimes between the impedance of the two outputs.
This makes sense so far...
This does in some cases causes color smear or phase shift.
You've lost me here - the Y cable is on a component out, which does not rely on phase shift for color information. If an impedance problem causes the levels to be different, it doesn't matter, because each Pb/Pr is affected in the same way. The only problem is the relative difference between Y+Px and Y itself. This can be corrected in software with a multiplication factor. Did you think I was talking about svideo?

tedkunich:

From the Connexant 878a datasheet:

Digitized video needs to be band-limited in order to avoid aliasing artifacts.
Because the Fusion 878A samples the video data at 8 × Fsc (over twice the
normal rate), no filtering is required at the input to the A/Ds. The analog video
needs to be band-limited to 14.32 MHz in NTSC and 17.73 MHz in PAL/SECAM
mode. Normal video signals do not require additional external filtering. After
digitalization, the samples are digitally low-pass filtered and then decimated to
4 × Fsc. The response of the digital low pass filter is illustrated in Figure 3-3. The
digital low pass filter provides the digital bandwidth reduction to limit the video
to 6 MHz. Several graphs show -3dB @ 6MHz. That's pretty good! And the flash ADC is directly on the chip, so even the cheapest boards have the same excellent specs.

I have no doubt that what you were saying is valid for purely analog circuits, but digital oversampling filters can be almost perfect right to the edge of operation. There is also a "sharpness' register in the chip to let even higher frequencies through. Read a review of the Creative X-Fi soundcard, you will see +-.1dB all the way to Nyquist.

As for your other question, please see the thread in general discussion.

[tedkunich]

rfmmars:


This makes sense so far...

You've lost me here - the Y cable is on a component out, which does not rely on phase shift for color information. If an impedance problem causes the levels to be different, it doesn't matter, because each Pb/Pr is affected in the same way. The only problem is the relative difference between Y+Px and Y itself. This can be corrected in software with a multiplication factor. Did you think I was talking about svideo?


I'm sure Richard was refering to impedance mismatches and other "strangness" that can happen when you tie two fairly broadband amplifiers together in an unitended manner. Depending on the circuit design, the change in impedance seen by the output amplifier or the input amplifiers could possibly affect the circuits response over its range of frequencies introducing phase delays, which could appear as color offsets or smearing.



From the Connexant 878a datasheet:

Several graphs show -3dB @ 6MHz. That's pretty good! And the flash ADC is directly on the chip, so even the cheapest boards have the same excellent specs.

I have no doubt that what you were saying is valid for purely analog circuits, but digital oversampling filters can be almost perfect right to the edge of operation. There is also a "sharpness' register in the chip to let even higher frequencies through. Read a review of the Creative X-Fi soundcard, you will see +-.1dB all the way to Nyquist.

As for your other question, please see the thread in general discussion.

The Conexant 878 is a next generation device that uses 2x oversampling and is earmarked for the HDTV market. I have assumed that in all of these discussions we have been refering to your ordinary BT848 or CX23881 devices - they are not quite in the same class :) I believe that you are confused by the 2x oversampling nomenclature - what they are doing is using a device that is capable of sampling at 2x faster than the highest frequency that they anticipate (i.e. input << Nyquist for that device) and then bandwidth limit the result in the digital domain. The Fusion 878 is sampling at 28MHz whilst your run of the mill BT848's (and co.) are sampling at 14MHz. This is no great revolution in technology, just that the ecomomies of scale allow for manufacturers now to do so at a price point at which they can make a profit. I'd be willing to bet that Conexant put a filter on each of the inputs to ensure no aliasing would be happening - the goal these days to make these devices idiot proof and minimize external components - that data sheet quote is misleading since the filtering MUST be there, it is just not external to the device.

Oh, BTW, the sharpness register on the BT chip sets only boosts gain of the higher frequencies by changing the taps in the bandpass filter - is does not change the cutoff, just the shape of the rolloff (think peaking just before rolloff)

T

[jmac698]

tedkunich: Depending on the circuit design, the change in impedance seen by the output amplifier or the input amplifiers could possibly affect the circuits response over its range of frequencies introducing phase delays, which could appear as color offsets or smearing.

Ok, that make sense. I can use a test pattern for relative chroma delay to test it, or even better, the same frequency sweep in Y and U/V to see if there's a difference when using the Y splitter. I did notice changes in signal strength, which was taken out by a simple brightness adjustment.
tedkunich: I have assumed that in all of these discussions we have been refering to your ordinary BT848 I have one I can test, and see what the freq. response is. Anyhow, in my test results I found >1.6 times the freq. response with my technique.
tedkunich: I believe that you are confused by the 2x oversampling nomenclature - what they are doing is using a device that is capable of sampling at 2x faster than the highest frequency that they anticipate (i.e. input << Nyquist for that device) and then bandwidth limit the result in the digital domain. I agree, why did you think I was confused? Because I said something about the sharpening? Actually, I didn't read the sharpening feature too closely, and you are probably right about the peaking. I understand about digital filters.
tedkunich: I'd be willing to bet that Conexant put a filter on each of the inputsA diagram shows an optional pi filter (I think), I could type the values into a program to see what kind of filter it is, but probably a lowpass somewhere between 7 and 14MHz. The implication is that if the optional circuit is shown, it's certainly not included in the chip. I don't see a problem with this, it's fairly valid to assume a signal is bandpass limited (in this case!). Also, with sigma-delta converters, filtering isn't used either usually (I'm thinking audio adcs here), because the 1-bit sample rate is so high. That's not the case here, though, it's just a standard flash converter.

Anyhow, did you look at my tests? I don't think cables cause my smearing problem. I'll have to take a picture to show you, and see if you can't replicate it.

I'm starting to think now that the svideo itself is mostly the problem, and maybe my outputs aren't 4:4:4 afterall, but it doesn't matter as there's a significant beneift to avoiding svideo anyhow.

[tedkunich]

Ok, that make sense. I can use a test pattern for relative chroma delay to test it, or even better, the same frequency sweep in Y and U/V to see if there's a difference when using the Y splitter. I did notice changes in signal strength, which was taken out by a simple brightness adjustment.
I have one I can test, and see what the freq. response is. Anyhow, in my test results I found >1.6 times the freq. response with my technique.
I agree, why did you think I was confused? Because I said something about the sharpening? Actually, I didn't read the sharpening feature too closely, and you are probably right about the peaking. I understand about digital filters.
A diagram shows an optional pi filter (I think), I could type the values into a program to see what kind of filter it is, but probably a lowpass somewhere between 7 and 14MHz. The implication is that if the optional circuit is shown, it's certainly not included in the chip. I don't see a problem with this, it's fairly valid to assume a signal is bandpass limited (in this case!). Also, with sigma-delta converters, filtering isn't used either usually (I'm thinking audio adcs here), because the 1-bit sample rate is so high. That's not the case here, though, it's just a standard flash converter.

Anyhow, did you look at my tests? I don't think cables cause my smearing problem. I'll have to take a picture to show you, and see if you can't replicate it.

I'm starting to think now that the svideo itself is mostly the problem, and maybe my outputs aren't 4:4:4 afterall, but it doesn't matter as there's a significant beneift to avoiding svideo anyhow.

I saw you pics, I guess that I did not quite understand some of them... I see that the svid was severely limited in B/W compared to the others, but I did not see any smearing on the others, was there supposed to be any? Also, what is the resolution that the bands on the right side represent? 4MHz? 6MHz? How was this signal generated?

T

[rfmmars]

How was this signal generated?



I would have liked to seen a final RBG frame (real video stuff) along with the sweeps.

Richard

[jmac698]

Hmm, you need to see, side by side, obvious difference comparison.

Technical measurements: still useful, however. Note the 6 bars at the top, these are MHz markers. If colors under bar4, it means the frequency is 4MHz. The svideo stopped before bar4, while the C4All continued to the very edge, which is 6MHz. I'll make labels.

For the b/w pictures, it's the same thing, but showing the direct color plane (U/Cb,Pb). This is pure color info translated to b/w. It's a little easier to see.

How was it generated? Don't know, it's from a reference test disc, have faith ;)

Smearing: not here. Example coming.

New tests coming.

My technical tests showed, 1) no freq. response difference btw svideo cables 2) svideo color response to 3.75MHz 3) C4All response off the scale, at least 6MHz.

[tedkunich]

Hmm, you need to see, side by side, obvious difference comparison.

Technical measurements: still useful, however. Note the 6 bars at the top, these are MHz markers. If colors under bar4, it means the frequency is 4MHz. The svideo stopped before bar4, while the C4All continued to the very edge, which is 6MHz. I'll make labels.

For the b/w pictures, it's the same thing, but showing the direct color plane (U/Cb,Pb). This is pure color info translated to b/w. It's a little easier to see.

How was it generated? Don't know, it's from a reference test disc, have faith ;)

Smearing: not here. Example coming.

New tests coming.

My technical tests showed, 1) no freq. response difference btw svideo cables 2) svideo color response to 3.75MHz 3) C4All response off the scale, at least 6MHz.


It would be interesting to see the schematic of the DVD player you are using. I wonder what type of filtering may be happening in the SVideo output circuit inside the player itself. I know that I was able to resolve 5.5MHz out of my Pioneer DVL-919 deck with my Winfast card (CX23881 based) using S-Video doing my LD captures.

T

[jmac698]

tedkunich: I know that I was able to resolve 5.5MHz out of my Pioneer DVL-919 deck with my Winfast card (CX23881 based) using S-Video doing my LD captures.
aha, very interesting, thus creates your strong impression of the clarity of svideo, compared to my poor impression. Tell me, how did you make your measurements, and you're sure it was the color part only, not the luma which is also 6MHz for me as well?

This is very interesting, some kind of professional outboard adc/dac
http://www.aja.com/iofamilyspecs.html
Video Output

IoLD: 8 or 10 bit SDI, SMPTE-259

IoLA:
Composite/S Video:
NTSC, NTSCJ, PAL
12 bit D/A, 8x oversampling
+/- .2 db to 5 MHz Y Frequency Response
+/- .2 db to 1 MHz C Frequency Response
.5% 2T pulse response
< 1% Diff Phase
< 1% Diff Gain
Component: SMPTE/EBU N10, Betacam 525 line, Betacam 525J, RGB
12 bit D/A, 8x oversampling
+/- .2 db to 5.5 MHz Y Frequency Response
+/- .2 db to 2.5 MHz C Frequency Response
.5% 2T pulse response
<1 ns Y/C delay inequity
I'm not sure what that means as far as relative quality of svideo vs component, and why I measure >1MHz.

[tedkunich]

aha, very interesting, thus creates your strong impression of the clarity of svideo, compared to my poor impression. Tell me, how did you make your measurements, and you're sure it was the color part only, not the luma which is also 6MHz for me as well?

OK, guess I need to clarify that a bit, capping with svid using DVD essentials, the resolution plate came through clearly - the plate however is only in B/W. The chroma goes out that far but I have never bothered to look to check the limits. I believe that the chroma on Svid is limited to around 1MHz or so. Back in the day when they created the NTSC standard, bandwidth was at a premium so they did a lot of tricks to minimize the amount of information that it carried, one of them was to limit the amount of color information as the eye is not that sensitive to fine detail in color. I believe that Svid interface prety much follows the NTSC specs and thus limits what is carried. No doubt that component out will carry more information.

http://www.ntsc-tv.com/images/tv/sub-csp.gif

You can see that the NTSC standard does not carry much high frequency information in the chroma. Check out this site for more info if you are so inclined... http://www.ntsc-tv.com The layout is really goofy but there is some good information in there.


T

[jmac698]

Have we been talking about different things all this time? When I said my resolution was 3.75MHz I meant the chroma part only. the luma part is 6MHz, and therefore, my C4All technique has chroma also at 6MHz.
The smearing I now believe is an inherent function of svideo, because the chroma information is lowpass filtered relative to the luma.

I posted some new tests, which show this smearing. You can't read the colored text in svideo, but you can in component.

But what is really going on here is quite complicated;
1) The chroma offset of YV12 sampling adds 1 pixel of blurriness around a color boundary,
2) The DVD player is upsampling the YV12 chroma with bicubic or higher order interpolation,
3) Svideo itself is filtering the resulting chroma

To really know my final chroma resolution measurement I think I need a controlled source like a video card output, with no chroma upsampling.
Anyhow, I'm quite convinced by now it gives at least 2x better quality. Aren't you?
To finish this off, I only have to workout the calibration issues, that are giving that stupid color cast.

[Qjimbo]

Wow pretty inventive! But surely you're limited to what things you can do this with? You wouldn't be able to do it from a live HD source because you have to cap it several times...

What'd be cooler is if someone found a way of putting RGB data straight into the Theater 550 :)

[jmac698]

Thank you :) Most people think it's a waste of time, but I think it's neat, and I'm doing it for fun.
Actually, I can record RGB, in fact even VGA output at the proper refresh rate could be capped. Also old systems like Amiga 2000 could be capped.
With TVTool, most Nvidia cards can output component or RGB data, with a special cable, also the Hollywood Plus can output component or RGB (I have the cable and have tested this). So I can test this next.
You can get even higher rez by capping with PAL-M, in black and white it's like NTSC but with 768 pixels by default.
As for live sources - yes that is a problem, but something good could be done with 2 or 3 capscards at the same time. A single PVR350 has dualtuner, this could be used. I would cap the Svideo and P+Pr signal, as your eye is more sensitive to color rez here. They could be merged in higher quality.
Potentially you could write a program to switch video inputs every field, because of 3:2 pulldown you could completely cap live, because the fields repeat and you would switch inputs.
Almost anything is possible :) Wow, you just gave me a bunch of ideas, thanks :)

[Qjimbo]

Actually, I can record RGB, in fact even VGA output at the proper refresh rate could be capped.
What card are you using? I know some cards like the Sweetspot can do RGB, but they're software only (and pricey). I reckon my Theater 550 card it must break down the composite/svideo into rgb at some point, though I could be wrong. If I'm right however all would be required would be to solder wires onto the board at this point in the circuit. The hard part is identifying where it does this exactly (if at all).

[jmac698]

Nope! Any card can record component, or RGB, with no hardware mods. But you do need a special cable.
Same as my component for all technique, just connect G+B together, connect to the svideo in (luma pins), cap. Connect G+R together, connect to the svideo in (luma pins), cap.
Then connect G by itself, cap.
Then a simple script will subtract (G+B)-G=B, (G+R)-G=R, and you have G. G must always be connected, because it has "sync pulses", and without those detected, the capcard will not activate.
There are no limits to recording anything, as long as you have a sync pulse with suitable timing that can be recognized in Pal or NTSC timings. VGA has separate Hsync and Vsyn pulse you could wire together, with a color, and there's a chance the capcard will pick it up.

[Qjimbo]

Oh I see what you mean, but it's hardly smooth because it has the same problems as your first method. I have a few games consoles that output as RGB, I was thinking of capping those, which wouldn't be possible using your multipass method...

One thing I did notice a while ago on the datasheet for the BT878A card was that it had 4 composite inputs, check it out here (http://www.olifantasia.com/pub/projects/gnuradio/doc/specsheets/bt8x8/Bt878A_Data_Sheet_100600B.pdf). Such a thing might be usable for your Component method, it's odd that these extra connectors aren't used anywhere.

EDIT: Seems that that is quite a high spec. chip, not your average BT chip. But anyway...

[jmac698]

If it were a repeatable thing, like an old Amiga demo, or the "attract mode" loop at the start of a game, then yes, but I see the real need for live capture here.
I thought of this before; what you need are 3 devices, like VCR, dvd recorder, and computer, or dual tuner cap card and VCR, etc. There is one problem still, the Y needs to be replicated 3 times; you can use a distribution amplifier for this, it only need to be 1 to 3 way composite. Then you can connect all combinations at the same time. I'm worried about how stable the VCR is, though. The dvd recorder can essentially replace one computer with a capcard.
Btw, you can also use two satellite boxes tuned to the same channel - what you get is "collaborative recording", where one person records part of it, someone another part, this can be live, but then you have to share files.

[jmac698]

It turns you can view my vyu mode live, on your TV right now, by changing one setting. Check your hidef source that has HDMI/DVI output. See if it has a setting to output RGB or YCC 4:4:4. Change RGB to YCC, and it's the same as my vyu, in other words the screen will look very purple. That's because R and B averages around 128 now.

[jmac698]

Bumping this because someone asked if it were possible to record component video.
As for news, nothing has changed due to the inability of avisynth to return the value of a pixel in script.
If the pixelinfo filter is ever finished, it will be possible to create a semi-automatic version of the "component4all" technique, in terms of the calibration proceedure.
Until then, again a summary:
It is possible for ANY capture card to record component video. The benefits of this is at least 1.6 better looking picture, as compared to an svideo picture, this was tested scientifically.
There is a disadvantage in that you need to record the exact video 3 times in a row; each pass is one color of the component.

[jmac698]

Just like to mention I wrote a real avisynth compiled filter for this. Might play again with it some day just for fun. I've also captured 480p, though there is a slight problem with some pixels cutoff horizontally. If capturing a pulled down movie, this effect can be compensated exactly, and for general video, you might have to fill in missing strip with standard def image...

[2Bdecided]

Wouldn't the ability to capture 480p depend entirely on your card's ability to find the half-width H-sync in a 480p signal? Some may, and transform two input 480p lines onto one output 480i line - but don't some just give up?

Also, given that 480p requires twice the bandwidth, and a 480i card doesn't have it, you're effectively sampling at 360 pixels per original line, rather than 720.

Still an interesting technique, and good to play around with for 480i/576i component. However, for real time capture, rather than buying 3 PCs and 3 standard capture cards, it would be cheaper to buy one PC and a component capture card.

Cheers,
David.