Video Color Space is a mystery to me..
May 15, 2006 3:22 PM Subscribe
I would like to find resources, either in print or online, for learning as much as I can about video color space, specifically where it applies to compositing/keying work.
I'm an assistant video editor, principally in avid systems, who's looking to learn what I can about the nature of color spaces for different video formats. For instance, I'm told that minidv, having a color space of 4:1:1, is worthless for compositing and keying work. DVCPRO, having a 4:2:2 color space, I'm told is much better suited to that work, and other formats are even better. But 4:1:1 doesn't mean anything to me, technically. I just remember "4:1:1 is bad for keying, and that's what minidv has, so minidv is bad for keying." I think it's time for me to understand what those numbers mean, and why certain numbers are better than others, and how they apply to compositing and editing scenarios that I will encounter in my work. thanks in advance.
I'm an assistant video editor, principally in avid systems, who's looking to learn what I can about the nature of color spaces for different video formats. For instance, I'm told that minidv, having a color space of 4:1:1, is worthless for compositing and keying work. DVCPRO, having a 4:2:2 color space, I'm told is much better suited to that work, and other formats are even better. But 4:1:1 doesn't mean anything to me, technically. I just remember "4:1:1 is bad for keying, and that's what minidv has, so minidv is bad for keying." I think it's time for me to understand what those numbers mean, and why certain numbers are better than others, and how they apply to compositing and editing scenarios that I will encounter in my work. thanks in advance.
Best answer: Have a look at the book "Digital Compositing for Film and Video" by Steve Wright from Focal Press. I think there's an updated version that came out this year. Its a bit hardcore technical but you'll come out of it with a fairly complete understanding of the underlying processes of any modern compositing app.
posted by HK10036 at 5:42 PM on May 15, 2006
posted by HK10036 at 5:42 PM on May 15, 2006
Best answer: Well, the wikipedia is a good followup to this (and I'm simplifing a bit...but it'll make sense.)
Video is NOT RGB space.
This is Color space before...well, color space (at least in video)
The problem occured when all there was in video was "Luma" or black and white (which we will call Y)
Now, you invent color TV process (well invent a way to add it to the signal), slow down the signal to 29.97 and well...you have a problem......
How do you keep broadcasting to the Black and white TVs?
By adding a Chroma Red (Cr) and a Chroma Blue (Cb) channel.
Where will we put the Green? 60% goes in the Y channlel, and 20% go in the U and V Channels respectfully. This gives us YCrCb (but easier to think of as YUV).
Beautiful! Think about how cool that solution is. Now we have one signal that has both BW and Color....and the color TVs just use the added color signal. Any color signal is ignored by BW tv.
Oops. The problem is our tape/broadcast formats can't handle all the samples. How can we lower it?
Well, the eye sees Black and White better than color.
Lets take 4 Luma Samples for every 2 Chroma samples
This gives us 4 Y (Luma + some green) samples for every 2 UV (Chroma) samples.
Samples? Sure.
NTSC has 720 horizontal samples (and 480 or 486 lines, depending on format).
We'll Sample the 720 on every pixel for luma (yielding 720 luma samples...don't forget the green is in there too.)
But for each color channel, we'll half that...c'mon, nobody will notice. So, we'll just take 360 samples (1/2 as many as the luma).
This yields the ratio for YUV of:
4:2:2 or
720 Luma (+60% green), 360 Chroma red (+ 20% green), 360 Chorma Blue (+20% green.)
Much of the color can be interpolated back...since we've only lost half.
AH, I hear what you're saying. You want a really cheap digital format. Right. Because if it's digital it can be smaller/cheaper/faster.
There's no way to get that data on that little tape. I bet we could compress it (a la JPEG) and throw away even more color. After all, nobody really complained when we threw away half. Let's do it again, so we only take 1/4 of the samples.
So, 4:1:1
720 Luma (+60% green), 180 Chroma red (+ 20% green), 180 Chorma Blue (+20% green.)
It gets interpolated back to 720...but it doesn't matter. Most of the color is gone.
And for green (or chroma) keying, since so much of the other channels is gone (and being interpolated)....realistically a so-so key in 4:2:2 space...is a shitty key in 4:1:1 space.
posted by filmgeek at 6:13 PM on May 15, 2006
Video is NOT RGB space.
This is Color space before...well, color space (at least in video)
The problem occured when all there was in video was "Luma" or black and white (which we will call Y)
Now, you invent color TV process (well invent a way to add it to the signal), slow down the signal to 29.97 and well...you have a problem......
How do you keep broadcasting to the Black and white TVs?
By adding a Chroma Red (Cr) and a Chroma Blue (Cb) channel.
Where will we put the Green? 60% goes in the Y channlel, and 20% go in the U and V Channels respectfully. This gives us YCrCb (but easier to think of as YUV).
Beautiful! Think about how cool that solution is. Now we have one signal that has both BW and Color....and the color TVs just use the added color signal. Any color signal is ignored by BW tv.
Oops. The problem is our tape/broadcast formats can't handle all the samples. How can we lower it?
Well, the eye sees Black and White better than color.
Lets take 4 Luma Samples for every 2 Chroma samples
This gives us 4 Y (Luma + some green) samples for every 2 UV (Chroma) samples.
Samples? Sure.
NTSC has 720 horizontal samples (and 480 or 486 lines, depending on format).
We'll Sample the 720 on every pixel for luma (yielding 720 luma samples...don't forget the green is in there too.)
But for each color channel, we'll half that...c'mon, nobody will notice. So, we'll just take 360 samples (1/2 as many as the luma).
This yields the ratio for YUV of:
4:2:2 or
720 Luma (+60% green), 360 Chroma red (+ 20% green), 360 Chorma Blue (+20% green.)
Much of the color can be interpolated back...since we've only lost half.
AH, I hear what you're saying. You want a really cheap digital format. Right. Because if it's digital it can be smaller/cheaper/faster.
There's no way to get that data on that little tape. I bet we could compress it (a la JPEG) and throw away even more color. After all, nobody really complained when we threw away half. Let's do it again, so we only take 1/4 of the samples.
So, 4:1:1
720 Luma (+60% green), 180 Chroma red (+ 20% green), 180 Chorma Blue (+20% green.)
It gets interpolated back to 720...but it doesn't matter. Most of the color is gone.
And for green (or chroma) keying, since so much of the other channels is gone (and being interpolated)....realistically a so-so key in 4:2:2 space...is a shitty key in 4:1:1 space.
posted by filmgeek at 6:13 PM on May 15, 2006
Best answer: One of the best articles on this subject is by the awesome Graeme Nattress, maker of many awesome-er FCP plugins, and is now on the development team of the mythical Red One camera
posted by melorama at 10:45 PM on May 15, 2006
posted by melorama at 10:45 PM on May 15, 2006
Best answer: oops...bad paste.
Here's the correct link:
http://www.lafcpug.org/Tutorials/basic_chroma_sample.html
posted by melorama at 10:55 PM on May 15, 2006
Here's the correct link:
http://www.lafcpug.org/Tutorials/basic_chroma_sample.html
posted by melorama at 10:55 PM on May 15, 2006
Response by poster: you know, I'm marking melorama's first answer as a best one, anyway, just because he called the Red One camera mythical and that amuses the hell out of me.
posted by shmegegge at 6:40 AM on May 16, 2006
posted by shmegegge at 6:40 AM on May 16, 2006
« Older Intractable math / physics puzzle: Light in a... | Good styles for good business site design? Newer »
This thread is closed to new comments.
posted by j at 4:51 PM on May 15, 2006