8bit 4:4:4 or 10bit 4:2:2 at DVinfo.net

Dek Rypter · July 10th, 2007, 12:19 PM

What's better for color grading/correction:
8bit 4:4:4 or 10bit 4:2:2 ?

Glenn Chan · July 10th, 2007, 02:01 PM

Probably 8-bit 4:4:4 R'G'B'.

Bit-depth-wise, you get some rounding error converting 10-bit Y'CbCr to 8-bit R'G'B'... so Y'CbCr does not use its bits that effectively.

There are also variations on how things are implemented...
full scale or studio range (0-255 range or 16-235 range if it's an 8-bit format)
resampling method used- many NLEs do this poorly
rounding behaviour

2- On the other hand, other factors are also important. You usually don't get a 8-bit 4:4:4 R'G'B' image to play with... most video cameras don't shoot that.

Thomas Smet · July 11th, 2007, 08:56 AM

It kind of depends on how much you will push the color.

10bit 4:2:2 has less banding because it has more color samples to work with. You can really push the color to extreme levels and still not get any banding. It is true however that YUV or YCC color isn't very accurate. Converting RGB (which is the way light works in nature) to YUV has to throw out a lot of color information.

8bit RGB 4:4:4 has accurate color but it could have banding if you push it very hard due to the limited amount of samples.

Usually I would also suggest 8bit RGB but like was said above chances are you will never see RGB video footage unless you are shooting with a SONY F950 and even then footage from the F950 is usually captured as 10bit RGB so it isn't the same as 8bit RGB. If you are working with computer generated images then of course I would keep it as 8bit RGB or use 16bit RGB still image sequences. If the material is computer generated then chances are you will not need to push the color very hard. You should just render it in the color or as close as you can so you do not have to push it very hard.

David Heath · July 11th, 2007, 09:46 AM

For colour GRADING (as the question asked) I'd expect 10 Bit to be more important than colour space. If we were talking about keying (say), 4:4:4 may have more relevance.

RGB may well represent the way light works in nature, but subsampled YUV may approximate better to the way the human eye sees. The theory is that it discards information that the eye is least able to make use of.

I suspect there are so many other factors to consider that there is no simple yes/no answer to the question as posed.

Dek Rypter · July 12th, 2007, 08:14 AM

Thanks for the information.

Since I discovered that HDV is fine for keying (my delivery format is SD) I do a lot of CGI with multiple layers to create artificial backgrounds. It takes too much time to re-render the background layers after making changes in the top-layers so I would like to export my backgrounds as AVI files and re-import them for futher processing. I guess uncompressed RGB 4:4:4 would be best for this.

Glenn, I read your exellent articles and maybe you or someone else can help me with the following. I'm lookin for the most efficient way to archive my HD masters and was thinking of this workflow:

- Edit native HDV (4:2:0) including transitions etc, but without grading
- Export as uncompressed RGB (4:4:4), about 22 GB for a 3 minute video
- Use VirtualDub to send 4:2:0 to Alparysoft lossless codec and end up with a lossless compressed AVI file of about 2 GB

Do I retain all information/quality with this method?

Glenn Chan · July 12th, 2007, 11:45 AM

I don't think a lossless codec will compress that much... they average to around 2:1 to 3:1.

2- With virtualDub everything passes through as 8-bit RGB usually... so to keep things mathematically lossless, use a lossless codec that operates on RGB. (Not ones that convert to so-called "YUV" and compress that.)

Dek Rypter · July 12th, 2007, 05:19 PM

I agree with the 2:1 or 3:1 compression ratio. HuffYuv resulted in a file of 10 GB. Alparysoft (in lossless mode) resulted in a 5 GB file (4:1 compression ratio). Virtualdub processes everything in RGB, but can send other colourspaces to the codec. I tried YV12 4:2:0 so the codec didn't have to do any colourspace conversion and this resulted in the "lossless" 2 GB AVI file (10:1 compression ratio). Should I be suspicious of the Alparysoft codec and stick with HuffYuv? I think that if HuffYuv could work with YV12 4:2:0 it would result in a file of around 4 GB.

If you convert 4:2:0 to 4:4:4 and back to 4:2:0 do you lose information if the codec is really lossless?

Gary Gowman · July 14th, 2007, 04:31 AM

10 bit

Gary

Glenn Chan · July 15th, 2007, 02:23 PM

Hmm I might be wrong about Virtualdub. The way to check would be to do a difference test.

2- RGB <--> YCbCr ("YUV") conversions are generally lossy.

Converting between chroma subsampled and non-chroma subsampled formats are generally lossy if done right.

The way to figure out would be to do a difference test... see if the numbers change by using a difference composite mode.

July 10th, 2007, 12:19 PM	#1
Dek Rypter Regular Crew Join Date: Mar 2007 Location: Rotterdam, The Netherlands Posts: 25	8bit 4:4:4 or 10bit 4:2:2 What's better for color grading/correction: 8bit 4:4:4 or 10bit 4:2:2 ?

July 10th, 2007, 02:01 PM	#2
Glenn Chan Inner Circle Join Date: Jun 2003 Location: Toronto, Canada Posts: 4,750	Probably 8-bit 4:4:4 R'G'B'. Bit-depth-wise, you get some rounding error converting 10-bit Y'CbCr to 8-bit R'G'B'... so Y'CbCr does not use its bits that effectively. There are also variations on how things are implemented... full scale or studio range (0-255 range or 16-235 range if it's an 8-bit format) resampling method used- many NLEs do this poorly rounding behaviour 2- On the other hand, other factors are also important. You usually don't get a 8-bit 4:4:4 R'G'B' image to play with... most video cameras don't shoot that.

July 11th, 2007, 08:56 AM	#3
Thomas Smet Trustee Join Date: Mar 2004 Location: Milwaukee, WI Posts: 1,719	It kind of depends on how much you will push the color. 10bit 4:2:2 has less banding because it has more color samples to work with. You can really push the color to extreme levels and still not get any banding. It is true however that YUV or YCC color isn't very accurate. Converting RGB (which is the way light works in nature) to YUV has to throw out a lot of color information. 8bit RGB 4:4:4 has accurate color but it could have banding if you push it very hard due to the limited amount of samples. Usually I would also suggest 8bit RGB but like was said above chances are you will never see RGB video footage unless you are shooting with a SONY F950 and even then footage from the F950 is usually captured as 10bit RGB so it isn't the same as 8bit RGB. If you are working with computer generated images then of course I would keep it as 8bit RGB or use 16bit RGB still image sequences. If the material is computer generated then chances are you will not need to push the color very hard. You should just render it in the color or as close as you can so you do not have to push it very hard.

July 11th, 2007, 09:46 AM	#4
David Heath Inner Circle Join Date: Jan 2006 Posts: 2,699	For colour GRADING (as the question asked) I'd expect 10 Bit to be more important than colour space. If we were talking about keying (say), 4:4:4 may have more relevance. RGB may well represent the way light works in nature, but subsampled YUV may approximate better to the way the human eye sees. The theory is that it discards information that the eye is least able to make use of. I suspect there are so many other factors to consider that there is no simple yes/no answer to the question as posed.

July 12th, 2007, 08:14 AM	#5
Dek Rypter Regular Crew Join Date: Mar 2007 Location: Rotterdam, The Netherlands Posts: 25	Thanks for the information. Since I discovered that HDV is fine for keying (my delivery format is SD) I do a lot of CGI with multiple layers to create artificial backgrounds. It takes too much time to re-render the background layers after making changes in the top-layers so I would like to export my backgrounds as AVI files and re-import them for futher processing. I guess uncompressed RGB 4:4:4 would be best for this. Glenn, I read your exellent articles and maybe you or someone else can help me with the following. I'm lookin for the most efficient way to archive my HD masters and was thinking of this workflow: - Edit native HDV (4:2:0) including transitions etc, but without grading - Export as uncompressed RGB (4:4:4), about 22 GB for a 3 minute video - Use VirtualDub to send 4:2:0 to Alparysoft lossless codec and end up with a lossless compressed AVI file of about 2 GB Do I retain all information/quality with this method?

July 12th, 2007, 11:45 AM	#6
Glenn Chan Inner Circle Join Date: Jun 2003 Location: Toronto, Canada Posts: 4,750	I don't think a lossless codec will compress that much... they average to around 2:1 to 3:1. 2- With virtualDub everything passes through as 8-bit RGB usually... so to keep things mathematically lossless, use a lossless codec that operates on RGB. (Not ones that convert to so-called "YUV" and compress that.)

July 12th, 2007, 05:19 PM	#7
Dek Rypter Regular Crew Join Date: Mar 2007 Location: Rotterdam, The Netherlands Posts: 25	I agree with the 2:1 or 3:1 compression ratio. HuffYuv resulted in a file of 10 GB. Alparysoft (in lossless mode) resulted in a 5 GB file (4:1 compression ratio). Virtualdub processes everything in RGB, but can send other colourspaces to the codec. I tried YV12 4:2:0 so the codec didn't have to do any colourspace conversion and this resulted in the "lossless" 2 GB AVI file (10:1 compression ratio). Should I be suspicious of the Alparysoft codec and stick with HuffYuv? I think that if HuffYuv could work with YV12 4:2:0 it would result in a file of around 4 GB. If you convert 4:2:0 to 4:4:4 and back to 4:2:0 do you lose information if the codec is really lossless?

July 14th, 2007, 04:31 AM	#8
Gary Gowman Regular Crew Join Date: Sep 2005 Posts: 47	10 bit Gary

July 15th, 2007, 02:23 PM	#9
Glenn Chan Inner Circle Join Date: Jun 2003 Location: Toronto, Canada Posts: 4,750	Hmm I might be wrong about Virtualdub. The way to check would be to do a difference test. 2- RGB <--> YCbCr ("YUV") conversions are generally lossy. Converting between chroma subsampled and non-chroma subsampled formats are generally lossy if done right. The way to figure out would be to do a difference test... see if the numbers change by using a difference composite mode.