View Single Post
Old 27th February 2006, 20:06   #6  |  Link
jmac698
Registered User
 
Join Date: Jan 2006
Posts: 1,867
debating...

Quote:
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.

Quote:
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

Last edited by jmac698; 27th February 2006 at 20:30. Reason: speling :)
jmac698 is offline   Reply With Quote