16 bit vs. 8 bit

[digitaldog]

What I seek would be an image that shows that 16 bit editing has a clear advantage, when working in a normal gamut RGB space, such as Adobe RGB.

What's a "normal" gamut, what's normal about Adobe RGB (1998) and what's an abnormal gamut?

[curvemeister]

...
I usually shoot HDR interiors by blending 3 bracketed images {-2,0,+2} corrected to initially have the same exposure as the least exposed shot, i.e. a very dark image. I then open it into PS and apply a lift curve, a contrast curve, and some local arrangements when needed
....
I know it's quite a extreme postprocessing, but just an example that I REALLY NEED 16 BIT for my interiors workflow.

BR

Thanks for an interesting example, BR.

Your work-flow strikes me as an adaptation of expose for the shadows, develop for the highlights approach.  Although you do not show it in your crop, there is important highlight detail and tonality in the ceiling lighting.  I think your image, in the context of your workflow, does make the case for 16 bit as a container for HDR processing.

It seems to me that, since your exposure covers 4 "bits" of dynamic range, it might be possible to get a similar banding effect in the shadows with a single shot, again staying in Adobe RGB.

[Guillermo Luijk]

Your work-flow strikes me as an adaptation of expose for the shadows, develop for the highlights approach.  Although you do not show it in your crop, there is important highlight detail and tonality in the ceiling lighting.  I think your image, in the context of your workflow, does make the case for 16 bit as a container for HDR processing.

This somewhat strange workflow is because I feel comfortable editing images straight from a neutral RAW development (i.e. all parameters set to 0), so I am not interested at all in all those sliders commercial RAW developers have to offer.
By doing so all images behave the same to me and I know what to expect. I can process them usually with two curves in Photoshop (bright and contrast) with little or no manual edition.

Actually the exposure is for the highlights: I do ETTR in the scene to preserve the highest lights I am interested in, and make that shot be the least exposed in the {-2,0,+2} bracketing. Then the blending program uses DCRAW for RAW development, and that means neutral output. The only way to preserve all DR captured and in a totally virgin image, is to adapt output exposure to the least exposed shot, so any information coming from the most exposed shots is corrected down in exposure.

Anyway the resulting image in 16 bit is so powerful that several copies of it can be produced at different overexposed values and then blended in any HDR program supporting TIFF files. It will not be fake HDR since all the DR was into the output TIFF.

BR

[jbrembat]

I have.

I usually shoot HDR interiors by blending 3 bracketed images {-2,0,+2} corrected to initially have the same exposure as the least exposed shot, i.e. a very dark image. I then open it into PS and apply a lift curve, a contrast curve, and some local arrangements when needed.

The initial image is obtained in 16-bit and Adobe RGB with 2.2 gamma using Zero Noise. That image needs to be very robust because shadows will be strongly lifted for tone mapping.

In 16-bit, these images are incredibly robust (I managed to lift here one of them by +12 EV, see Fig. 11, and there was no posterization. Not even a 16-bit linear DNG would have resisted such an overexposure, see Fig. 12). The trick for that is that both exposure correction and 2.2 gamma are applied at the same time, in just one high precission floating point operation, then rounded to 16-bit integer (BTW real 16-bit, not 15-bit like in Adobe software), so it's impossible to have a richer image for that degree of exposure.

On the contrary in 8-bit, the resulting image shows clear posterization in the shadows after the tone mapping process in PS. These images are almost noise free thanks to the optimum bracketed blending, so shadow posterization becomes easily visible on a low bitdepth:

Sample scene:


16-bit vs 8-bit postprocessing (posterization in the wooden table left and chair back when initial image was 8-bit):


For those interested in looking at the curves applied: capas.tif.

I know it's quite a extreme postprocessing, but just an example that I REALLY NEED 16 BIT for my interiors workflow.

BR

HDR.
Yes I know it, floating point not integers (16 bit)!
After the tonal compression you can go for 8 bit. That is what I do, but the same is true for any researcher I know (Reinhard, Fattal, Drago, Ward,Tumblin,.... and so on),

But just to avoid confusion: 8 vs 16 bit is the image bit depth. Depending on algorithms the internal computations  may be done with different numeric precision.
Just to give some example:
- going to CIE_Lab or CIE_XYZ for white balance is well performed using floating point (I use double precision!).
- transformations between color gamuts is internally performed in double (you have to go to/from XYZ or Lab, the PCS)

Jacopo

[jtrujillo]

Guillermo, interesting point focusing on shadows that need some heavy processing. I must say though that I have problems to perceive the differences in the crops you show.

Another experiment, I am not sure what point will support, it demonstrates that there's difference but I am not sure if that difference is perceivable. Anyway this case it is a single, very ordinary operation.

From this image:

Loaded two copies of that same image one in 8 and the other at 16 bit mode. I applied Unsharp mask 300 / 0.3 / 4 to both.

Overlapped one on top of the other as a layer and put the layer mode to difference. Everything black.
But on top I've placed a Levels layer and moved the white point to the left, almost to the location of black. The result:



This shows that there's significant information different between both images, you can recognize an image there!.
As I said I am not sure if it's perceivable at least consciously but I wonder if it might affect perhaps the outcome of a print?

[Jonathan Wienke]

If you work in ProPhoto RGB, or another wide gamut space, there is a decided need for 16 bit.  

In other color spaces, such as Adobe RGB or sRGB, you are very unlikely to see a difference.  Personally I'm still waiting for a convincing example.  There are any number of excellent photographers who do not feel the need for an example, and decide to work in 16 bit simply to play it safe.

Try http://www.visual-vacations.com/Photography/16_vs_8.htm

[curvemeister]

Try http://www.visual-vacations.com/Photography/16_vs_8.htm

One of the reasons that 8 bit works so well is gamma encoding.  Take that away, and there are a host of examples where 8 bit is inferior to 16 bit.  The example you refer to, being a linear encoded image, shows the importance of this.

[Jonathan Wienke]

The benefit of 16-bit workflow becomes obvious when applied to any image where aggressive level or curve adjustments are needed, not just linear-to-gamma-encoded tone curve conversion. Have you ever tried editing an underexposed JPEG vs a similarly underexposed RAW image? The shadow artifacts you'll find in the JPEG image vs the RAW are similar to those shown in my example. The same principle is true where the camera can barely capture all of the DR of the subject; working the image in 16-bit mode will give you cleaner shadow detail than an 8-bit version. The more aggressive the adjustment an image needs, the more beneficial 16-bit editing becomes.

Here's another example:
8-bit image file
16-bit image file

Process the 16-bit image in your usual manner, recording what you do to an action. The only rule is that you must stay in 16-bit mode from start to finish.
Run the action you just recorded on the 8-bit image, so that all steps are performed in 8-bit mode.
Compare the images side-by-side, and post the results.

Which image has the best shadow detail?