-
Chroma Subsampling in HDSLRs once and for all…
Hi everyone,
There seems to have been some debate for a while about a few elements of the Canon DSLR compression, especially chroma subsampling. First of all, I’d like to clarify that this is different to colour space which refers to how much of the gamut is seen. (Is that correct?).Secondly, I believe the H264 Canon DSLRs records to is subsampled to 4:2:0, NOT 4:2:2. Is that also correct?
😀
Tom Morter-Laing
Freelance Editor
Certified Apple Product Proffessional, 2010
http://www.depictproductions.co.ukSony Z5, with Rode NTG2.
iMac 27″ intel i7 2.93GHz, 12GB RAM, ATI HD5750 [1GB GDDR5], 2TB Int. SATA with 2TB External HDD; (FW800), with Elgato Turbo H264HD. -
10 Replies
-
It is 4:2:0
You should look into the cineform codec, for high end FX plates.*Production*Post-Production*
Apple Certified Instructor (Final Cut Pro 7)
“Semper Fi USMC” -
[Thomas Morter-Laing] “I’d like to clarify that this is different to colour space which refers to how much of the gamut is seen. (Is that correct?)”
Correct. Some times you’ll see color space incorrectly stated as Y’UV although this is technically incorrect for component digital video, Y’CbCr (which is used for digital video ) encoding of luma and chroma are separate: Y’ is gamma weighted luma and CbCr are the chroma difference channels. Because of this separation chroma subsampling can into play to provide further compression ability.
Because our eyes are much more sensitive to luma (brightness) the Y’ component is hardly ever reduced (a notable exception is HDCAM (not SR) which is 3:1:1) however chroma often is often reduced severely. 4:2:2 means that chroma is being sampled half as often as luma 4:1:1 a quarter as often and 4:2:0 color is sampled 4:2:2/4:0:0 alternating lines
What makes matters a little more confusing is REC 709. For the purposes of color REC 709 describes a gamut. That gamut for REC 709 is significantly smaller then what Y’CbCr is really capable of.
Robbie Carman
—————-
Colorist and Author
Check out my new Books:
Video Made on a Mac
Apple Pro Training Series DVDSP
From Still To Motion -
Robbie, thanks for this post, should hopefully help a few people because there still seems to be a lot of confusion.
Anthony, thanks for the tip but I’m not entirely sure what he cineform codec has to do with this post… it’s a good codec but not really what I was asking about, or have I missed something?Also, has anyone therefore found a way to prevent subsampling on any of the canon range? Im aware of the Magic Lantern firmware hacks, but they give you data rate control rather than anything else. Even the 5D MK2 is 4:2:0, which is a shame because I was going to get one with the intention of chroma keying work. I currently have a t2i (550d) which I love, but still not great for keying for this reason. (Yes I know about good lighting bla bla bla) 😀
Also people have mentioned someone called Syndicate? Where can I find them? I Googled it and found nothing…
😀
Tom Morter-Laing
Freelance Editor
Certified Apple Product Proffessional, 2010
http://www.depictproductions.co.ukSony Z5, with Rode NTG2.
iMac 27″ intel i7 2.93GHz, 12GB RAM, ATI HD5750 [1GB GDDR5], 2TB Int. SATA with 2TB External HDD; (FW800), with Elgato Turbo H264HD. -
[Robbie Carman] ” 4:1:1 a quarter as often and 4:2:0 color is sampled 4:2:2/4:0:0 alternating lines”
Robbie,
tis is not exactly like that.
4.1.1. uses blocks of 4 pixels in a line: 4 Luma samples + 1 Chroma sample4.2.0: Uses blocks of 4 pixels but in 2 lines , and two in the line below (2×2): 4 luma samples + 1 Chroma sample.
In the end 4.1.1 and 420 have the same number of Chroma samples per block.
Both have 25% of the original Chroma samples (4.2.2 has 50%).
The advantages or 4.2.0 over 4.1.1 are clear when we think about “blockiness”.
Here there are some graphics:
https://www.kenstone.net/fcp_homepage/chroma_investigation_nattress.html
rafael -
Rafael – we’re saying the same thing about 4:1:1 as far as 4:2:0 you are technically correct about its still being 25% of the original chroma. However 4:2:0 is a little more complex as there is variation on the vertical and horizontal sampling depending on format. However I agree that my 4:2:2/4:0:0 way of explaining it might be a little to “simple”
Robbie Carman
—————-
Colorist and Author
Check out my new Books:
Video Made on a Mac
Apple Pro Training Series DVDSP
From Still To Motion -
[Robbie Carman] “Rafael – we’re saying the same thing about 4:1:1 as far as 4:2:0 you are technically correct about its still being 25% of the original chroma. However 4:2:0 is a little more complex as there is variation on the vertical and horizontal sampling depending on format. However I agree that my 4:2:2/4:0:0 way of explaining it might be a little to “simple””
You are right Robbie,
Sorry I wanted to explain that difference, but in the end is what I forgot.
On 4.2.0, Cb and Cr are not coming from the same pixel.
One is sample from a line and the other from the line underneath.
That is on PAL and HD.
As you points the 420 on MPG2-2 has some differences.
rafael -
Canon has a new codec on there video cameras that is 4:2:2 mpeg, let’s hope it will find its way to the next round of DSLRs or that they will give clean HD video out on the HDMI ports.
-
[Michael Sacci] “Canon has a new codec on there video cameras that is 4:2:2 mpeg, let’s hope it will find its way to the next round of DSLRs or that they will give clean HD video out on the HDMI ports.”
Right. Canon is using SONY’s XDCAM-HD 422 at 50Mbps.
rafael -
Didn’t know it was just the Sony codec, would be a step is the right direction. Clean HD 1080p (or psf) would be great also.
Will the up and coming Sony S3 use that codec or is it using a 4:2:0 version?
-
Hi Michael,
I guess you mean the “Sony PMW F3”: a 18K camera with 444 options, and records 35Mbps 4.2.0.
For back-up is OK.
rafael
PS: SONY and PANA has the best technitians and developers, but when they have to put that technology together to be sold, they become I kind of “Jekyll and Mr Hyde”.
Reply to this Discussion! Login or Sign Up