Banding when doing gaussian blur (or iris blur) in 16-bit black and white mode

[Hjorthmedh]

Hej! 

I use the new Photoshop 25.1.0 and have a problem with gaussian blur and iris blur (it might also apply to other types of blur). I see the effect for example when editing 16-bit black and white images, and applying a blur with high amounts (say 100 pixels or 200 pixels). The blurred image has banding, the grey tones are not smooth as expected. If I instead switch to RGB 16-bit, I do not see the effect on a similar image.

It should be pretty easy to reproduce, take a black and white 16-bit image (in my case it had a bit of noise), add gaussian blur with 200ish in radius. You should be able to see the banding in the preview. Turn off the preview, banding is gone, turn it back on banding again. Switch to 16-bit colour, apply the blur, no banding.

I edit a lot of black and white images, so this is a big issue.

[Jeff Arola]

In the Adobe Photoshop 2024>Settings (Preferences)>Technology Previews do you have Precise previews for 16- bit documents checked?

[Hjorthmedh]

Hej Jeff, thanks for the quick reply. Yes the "precise previews for 16-bit document" is checked. The issue also persists if I apply the blur effect to the image and view it outside of preview mode.

[D Fosse]

With 16 bit data, any banding you see is in your display system.

Most monitors are 8 bit, some even 6 bit + temporal dithering. That means 256 discrete steps, and gamma encoding means that these steps will be further apart in the shadows. There is nothing you can do about that.

The only way to avoid it is to use a (expensive) 10 bit capable display. To not see it, add a tiny bit of noise to break it up.

Banding is cumulative. It will be exaggerated by calibration tables loaded from the video card (as opposed to in the monitor's own native processor) or a bad monitor profile.

[Hjorthmedh]

Hej Fosse,

I use a Macbook Pro M1 laptop, did a quick search and that should be a 10-bit capable display.

Also important to note is that if I do the blur when the same image is in 16-bit colour mode, there is no banding visible (so the display should be working).

In both cases the image looks black and white, since there is no colour in the actual image, but for some reason the fact that it processes in black and white seems to add banding.

[D Fosse]

A 10 bit display is an Eizo Coloredge, NEC Spectraview and a few other budget models. Not a laptop.

EDIT: Re-reading your original post, I see you have this banding on 8 bit files but not 16 bit. So then everything is working as it should. There will be the 256 discrete steps, but under normal circumstances you won't see that.

8 bit data + 8 bit display is a cumulative effect.

[Hjorthmedh]

I think you missunderstand me. 

16-bit black and white --> problem with banding with iris blur and gaussian blur.

16-bit colour --> no problem with banding.

The images are in 16-bit mode, opened from 14-bit Nikon NEF files. The difference is in "black and white" mode vs colour.

[D Fosse]

Ah, sorry. That's what speed-reading does... 😉 You mean you are using grayscale mode, single channel?

In that case I can think of two things. One, that the display profile is a bit off-spec and doesn't handle grayscale input well. A common problem with grayscale is that it's very poorly supported anywhere outside Photoshop.

Another thing is that the grayscale document profile matters. The different grayscale profiles have very different tone curves, and the dot gain 10% and 15% profiles tend to push shadow values to much lower numbers. For a number of reasons this will exaggerate banding.

So try to convert to a grayscale profile that has the same tone curve as your RGB profile, and check again:

sRGB = sGray

Display P3 = sGray

Adobe RGB = Gray Gamma 2.2

ProPhoto = Gray Gamma 1.8

The dot gain profiles are in any case obsolete and shouldn't be used. For a generally safer default, set Gray Gamma 2.2. Photoshop still has the very unfortunate default of Dot Gain 15 %, which keeps getting people into trouble.

[Hjorthmedh]

Thanks Fosse, this is quite interesting and I am learning things.

I wrote "black and white", but it should be "grayscale" mode.

The profile was defaulted to “Working Gray - Dot gain 20%” for grayscale, I changed it to “Gray Gamma 2.2” (16-bit) and still see the same banding when doing Gaussian blur.

Undid the Gaussian blur, switched to RGB 16-bit, applied the same filter again and got a smooth gradient.

Also repeated it with 32-bit just to make sure it has enough bit-depth, but got the same result -- banding for grayscale, smooth gradient for colour.

[D Fosse]

OK. Unless it's the monitor profile, I'm out of options for the moment. But it could well be the monitor profile.

I just checked here to rule out any genuine Photoshop bugs, and I don't see any banding. But I do have 10 bit monitors.

[Hjorthmedh]

Fosse, your last reply got me thinking.

If the file data is OK, but the display is wrong, then blurring the 16-bit BW image (has banding), and then switching to RGB colour, should remove the banding in the image -- and it does! Before I had always undone the gaussian blur, switch to RGB, and then do the blur again.

That gave me hope, what if I converted the BW 8-bit jpeg to colour, would the banding disappear -- which it did not. So somehow the banding gets locked into the jpeg image (I use Adobe RGB).

What is the best grayscale profile to use? The Gray Gamma 2.2 gave banding, is there some wider profile? I feel like I am opening Pandora's box with all these different colour profiles...

[D Fosse]

It really sounds like your monitor profile isn't handling grayscale input well. Are you using a calibrator to make it? If so, changing some of the parameters and policies might fix it.

I usually advise people to avoid grayscale if possible, and the reason is that grayscale support is so poor everywhere. Very often the actual profile is ignored and the data are treated randomly and unpredictably. Nor is there a great incentive to fix bugs (as this may be).

Generally, gamma 2.2 is considered the safest, but no guarantees.

[Hjorthmedh]

I think you are right in that the grayscale profile is a problem, but I am not sure it is the only problem.

Doing blur on a grayscale 16-bit image (banding) and then converting it to RGB (looks fine).

If I save a grayscale image in 8-bit jpeg, I see banding there. But if the banding would only be in the "display step" (ie not in the image data itself), then opening the jpeg and saving it as a colour 8-bit image should remove the apparent banding (since we are now bypassing the grayscale profile), but it does not fix the issue.

It can still be the profile, but I think it is deeper than just in the step of how the picture is shown on screen, since the banding gets burned into a jpeg when saving it.

I will be doing more in RGB mode going forward and hoping that Photoshop adresses this in future updates. Really appreciate all the info and discussion. If anyone else has experience with this it would be interesting to hear.

[D Fosse]

There are two profiles involved here: the document profile and the monitor profile. They are both standard icc profiles, but they are separate and serve different purposes. Both need to be present and correct.

I don't think the document profile is a concern here, that's probably fine. I think the monitor profile is the problem. And that's why I asked where it came from - if you have made it with a calibrator, based on measuring your display, or if you're just using a generic/default system profile. Either can be bad.

When you display an image on screen, the numbers are converted from the document profile and into the monitor profile. Those recalculated numbers are sent to screen. This is a standard profile conversion, just like any other profile conversion, but executed on the fly as you work.

While it all ends up in the same destination profile, the source profiles are different. So the conversion is different; the actual math and tables are different. That's why it can work from, say, Adobe RGB, but fail from, say, ProPhoto RGB or Gray Gamma 2.2 or anything else.