16 bit vs 8 bit (one more time, sorry)

BH
Posted By
Bill Hilton
Oct 27, 2005
Views
441
Replies
6
Status
Closed
Someone posted this link on another NG from Deke McClelland ("Photoshop Bible" fame) which shows an image edited with only two operations (Levels and Auto Color) in both 8 and 16 bit, with a very noticeable difference. I know the lack of concrete examples is Mike’s main complaint about 16 bit advocates so I thought I pass this along …

http://www.graphics.com/modules.phpname=Sections&op=view article&artid=279

Just fyi as I think minds are already made up on both sides of the argument, and admittedly his first move in Levels is pretty radical …

Bill

How to Improve Photoshop Performance

Learn how to optimize Photoshop for maximum speed, troubleshoot common issues, and keep your projects organized so that you can work faster than ever before!

TA
Timo Autiokari
Oct 27, 2005
"Bill Hilton" wrote:

http://www.graphics.com/modules.phpname=Sections&op=view article&artid=279
admittedly his first move in Levels is pretty radical …

E.g. with an underexposed digicamera shot you do need pretty radical Levels adjustment. With scanned slide film you need pretty radical Curves adjustment. With scanned negative you need horrible Curves and Levels adjustments.

In addition, nearly all editing operations do add quantization noise (half of the least significant bit). This noise does accumulate (not directly additively, but it does accumulate) so after many editing operations there will be plenty of quantization noise in the image, easily up to two or even three lowermost bits.

In the 16-bit/c space (that in Photoshop only has 15 bits) this quantization noise does not matter much, 3 lowermost bits full of quantization noise still leave 12 bits for the image data.

In the 8-bit/c the 3 lowermost bits full of quantization noise leave only 5 bits for the image data.

Btw , in photographic workflow one does not need CS or CS2 for 16bit/c editing, I still use version 7 and fully 16-bit/c work is very easy with it.

Timo Autiokari http://www.aim-dtp.net/
T
toby
Oct 27, 2005
Bill Hilton wrote:
Someone posted this link on another NG from Deke McClelland ("Photoshop Bible" fame) which shows an image edited with only two operations (Levels and Auto Color) in both 8 and 16 bit, with a very noticeable difference. I know the lack of concrete examples is Mike’s main complaint about 16 bit advocates

There is no debate here. Photoshop provides 8 bit and 15 bit operations; those who care to can use either.

The *real* problem, and the usual cause for most postings here and in other forums, are that image processing newbies find the choice confusing: They don’t know when or why to use one or the other. THAT is the problem.

There is no "answer" to the question "which is better? 8 or 16?" It’s the same kind of useless timewaster as all the other Usenet perennials (vi or emacs? Linux or BSD? MySQL or PostgreSQL?) I can’t believe busy and intelligent people put energy into trying to answer unanswerables. As always: "use what works for you" and "it depends what you’re trying to do".

–T

so I thought I pass this along …

http://www.graphics.com/modules.phpname=Sections&op=view article&artid=279
Just fyi as I think minds are already made up on both sides of the argument, and admittedly his first move in Levels is pretty radical …
Bill
MR
Mike Russell
Oct 30, 2005
"Bill Hilton" wrote in message
Someone posted this link on another NG from Deke McClelland ("Photoshop Bible" fame) which shows an image edited with only two operations (Levels and Auto Color) in both 8 and 16 bit, with a very noticeable difference. I know the lack of concrete examples is Mike’s main complaint about 16 bit advocates so I thought I pass this along …
http://www.graphics.com/modules.phpname=Sections&op=view article&artid=279

This is an interesting example, but why didn’t Deke McClelland start with a normal photograph instead of one that has been heavily modified? I suggest it is because no such photograph exists.

The article demonstrates that 16 bits can represent more brightness levels than 8. It does not address the issue of whether any actual photograph looks better after editing in 16 bits per channel than in 8 bits.

Just fyi as I think minds are already made up on both sides of the argument,

Many minds are made up, and I respect those who have determined that 16 bits per channel, or 32 bits, is needed for them to achieve the desired result. I believe this issue is worth revisiting every once in a while, particularly when it is discussed constructively as has been the case here.

and admittedly his first move in Levels is pretty radical .

Yes, it is a radical levels move that, in effect, closes the image like an accordion enough to cause banding in 8 bit, but not 16 bit. The same sort of move, just a more extreme, could be used to show that 24 bits per channel is better for editing than 16, or that HDR (32 bits per channel) is better than 24. In each case, nothing practical is being demonstrated except that more bits equals more levels.

Once again, a qualification. I’m referring to color images in one of the customary Photoshop color spaces, specifically not a wide gamut space such as ProPhoto RGB. For those sorts of images, 8 bits per channel has plenty of head room even for extreme edits, and I have not yet seen such an image that shows otherwise.

Mike Russell
www.curvemeister.com
T
toby
Oct 30, 2005
Mike Russell wrote:

and admittedly his first move in Levels is pretty radical .

Yes, it is a radical levels move that, in effect, closes the image like an accordion enough to cause banding in 8 bit, but not 16 bit.

Dithering the output of the first adjustment would help a lot here, in 8 bit. Does Photoshop not do this?

But this is a silly example, since anyone with half a clue would understand that they had thrown away most of the significant bits by collapsing Levels in the first place (that can’t be retrieved by expanding it again).

–T

The same sort
of move, just a more extreme, could be used to show that 24 bits per channel is better for editing than 16, or that HDR (32 bits per channel) is better than 24. In each case, nothing practical is being demonstrated except that more bits equals more levels.


Mike Russell
www.curvemeister.com
MR
Mike Russell
Oct 31, 2005
"toby" wrote in message
[re banding showing up after a levels operation]
Dithering the output of the first adjustment would help a lot here, in 8 bit. Does Photoshop not do this?

Afaik there is no the only cases where Photoshop dithers is in gradients, profile conversions, and conversion from 16 to 8 bit mode. The first two are optional, and the third is always done. BTW – "dithering" is the adding of a typically small random value to break up banding patterns.

[re article discussing editing in 16 bits]
But this is a silly example, since anyone with half a clue would understand that they had thrown away most of the significant bits by collapsing Levels in the first place (that can’t be retrieved by expanding it again).

True enough. The scenario is unrealistic, and once again an artificial example – in this case a highly compressed photograph – is used to illustrate the superiority of editing in hibit.

It’s worth mentioning that there are any number of situations, ones that do not involve editing, where more than 8 bits per channel is useful. For instance a raw file may be underexposed several stops and still yield a relatively clean image. Raw files (which are linear gamma, BTW) can provide a handy way to multiply your ISO number by 8. If the same image is captured in 8 bits, such recovery is not possible.

Mike Russell
www.curvemeister.com
T
toby
Oct 31, 2005
Mike Russell wrote:
"toby" wrote in message
[re banding showing up after a levels operation]
Dithering the output of the first adjustment would help a lot here, in 8 bit. Does Photoshop not do this?

Afaik there is no the only cases where Photoshop dithers is in gradients,

Yes, a good feature. Before they did that, we used to have to do it by hand: Make gradient over full range, add appropriate uniform noise, then shrink levels to final range… The only way to avoid banding in small-range gradients. There was no corresponding workaround for banding in strong blurs (and I guess there still isn’t).

profile conversions, and conversion from 16 to 8 bit mode. The first two are optional, and the third is always done. BTW – "dithering" is the adding of a typically small random value to break up banding patterns.

….or error diffusion, which is arguably even more effective, since the resulting noise is subjectively less obvious. Looks like Photoshop could do with an option to dither adjustments, at least.

[re article discussing editing in 16 bits]
But this is a silly example, since anyone with half a clue would understand that they had thrown away most of the significant bits by collapsing Levels in the first place (that can’t be retrieved by expanding it again).

True enough. The scenario is unrealistic, and once again an artificial example – in this case a highly compressed photograph – is used to illustrate the superiority of editing in hibit.

Yes, I could add that anyone who did not realise they were throwing away data in 8 bits, would not make the connection that "16"-bit mode would help here.

It’s worth mentioning that there are any number of situations, ones that do not involve editing, where more than 8 bits per channel is useful. … —
Mike Russell
www.curvemeister.com

How to Improve Photoshop Performance

Learn how to optimize Photoshop for maximum speed, troubleshoot common issues, and keep your projects organized so that you can work faster than ever before!

Related Discussion Topics

Nice and short text about related topics in discussion sections