Curves Adjustment Layer vs direct Curves Image Adjustment results in different visible image

[Ian22444664yaqi]

I'm using Photoshop 2022 to stretch a 16 bit astrophotograph image utilizing multiple Curves Adjustment Layers. 

I noticed that when viewing the image at less than 100% zoom I see color banding and the histogram exhibits "combing" (I can refresh the histogram to remove the combing) following a single initial Curves Adjustment Layer operation.  The color banding only goes away if I zoom in to 100%+ (temporary only while zoomed in) or if I create a Stamp Visible layer (permanent).  This is not ideal because every time I add a Stamp Visible layer the file size subsantially increases.

If I instead apply the Curve directly to the image without an Adjustment Layer - I do not see the color banding or histogram combing.  The downside is that I am not able to modify the operation at a later time because I am not using adjustment layers.

I also noticed that this banding "issue" is not present when stretching a 32 bit image, only a 16 bit image.  Is this due to how Photoshop handles preview data?

[D Fosse]

In a word, yes.

Actually the tipping point isn't 100%, but 66.7 %. At lower zoom ratios, all previews are rendered at 8 bit depth. So the banding you see is 8 bit banding.

This is all part of a bigger picture to improve Photoshop performance. Another manifestation of this is that adjustment/blending previews are calculated based on the on-screen image, in other words downsampled at less than 100 % zoom. With very noisy and binary images, this can lead to misleading previews.

Bottom line - make it a habit to always check adjustments at 100 %. The significance of 100 % is that it represents one image pixel by exactly one screen pixel, and at full bit depth. It should be muscle memory - when in doubt, press ctrl+1.

At 100 % you will never see this discrepancy. It's a complete and faithful representation of the full image data.

[rayek.elfin]

In a word, yes.

Actually the tipping point isn't 100%, but 66.7 %. At lower zoom ratios, all previews are rendered at 8 bit depth. So the banding you see is 8 bit banding.

This is all part of a bigger picture to improve Photoshop performance.

By @D Fosse

It was indeed introduced a LONG time ago to improve Photoshop's performance - around 1992 at a time when PC and Mac computers were struggling to process 16 bit data, both from a CPU processing as well as memory requirements point of view.

Another decision at the time was to limit Photoshop's 16 bit mode to 15bit+1 to effectively halve the data overhead (32768 + 1 = 32769 in the (0,32768) range) which also simplified calculations. It also didn't matter for images at the time, because that full range wasn't ever used anyway.

But times have changed, of course.

Unfortunately this legacy code that was never updated and Photoshop's 16bit mode never received an upgrade to a proper full 16bit one. It is problematic, because in a time with full range HDR imagery used in various industries and for a wide range of applications, it means that when a full range 16bit data set image is loaded in Photoshop, effectively half the data is thrown away when the user saves a 16bit PSD or TIFF.

For example. I create a 65536px wide 16bit image in any other image editor (PhotoLine, Affinity, Krita, etc). I used PhotoLine to create a 65536 by 100px 16bit file. Then I create a grayscale gradient from pure black to pure white from left to right.

Then we zoom in and use the eye dropper to check the values. As expected: with each pixel the values drop or increase by 1, within a range from 0 up to 65336: a full 16bit value range. I save this as a tiff file.

I open the tiff file in Photoshop, and save the tiff as a new version. I check this file in PhotoLine, and low and behold: Photoshop downsampled the bit range. Each column of 2 pixels now shares the same value instead.

That means that in the original file I see a specific value for each pixel column: 22374, 22375, 22376, 22377, whereas in the Photoshop "16bit" image we see 22374, 22374, 22376, 22376 for the same 4 pixels.

This, of course, is completely unacceptable to anyone who needs the original value range to be kept intact. In particular for scientific image data research or HDR imagery used in film or 3d, it means that Photoshop runs amok with half your data -- and doesn't tell or inform the user. And that is really the worst part of it. It doesn't warn the user that this is happening, which may utterly throw off your calculations and results.

If you are doing astrophotography, this may also be problematic. Which is why anyone half-serious about 16bit image processing should always avoid Photoshop's half-baked "16-bit" mode. And all of this is caused by a legacy code base that is in dire need for an update to reflect the requirements and expectations of today's image processing.

The same holds true for the 16bit viewport downsampling to an 8bit range image pyramid when zooming out: it was implemented as a way to save on processing and (video?) memory resources back in the early 90s. But hello! The 21st century is calling, and these limitations should have been updated and fixed at least two decades ago.

NO OTHER modern image editor worth its salt destructively reduces 16bit image data like Photoshop does. NO OTHER moden image editor reduces the viewport preview quality of 16bit images either.

The only reason I can come up with why the Photoshop dev team hasn't already implemented a proper 16bit image mode, is that this legacy code is rooted and intertwined in many core areas of PS's code. It would mean a major rewrite, which they seem reluctant to do for whichever reason(s).

In any case @Ian22444664yaqi : depending on your astrophotography's 16bit data, I would hesitate to use Photoshop for this task, because you run the risk of losing half your original data. Or at the very least only work in 32bit mode, which behaves as it should. But never save or convert your 32 bit data as 16 bit in Photoshop unless you don't care about throwing away half of your 16bit data.

[Ian22444664yaqi]

@rayek.elfin @D Fosse 

Thank you both for the detailed explanations.  I now see the 66.7% threshold. 

I'm also going to take a closer look at the data loss issue that you noted when saving a 16bit PSD or TIFF file.  It is unfortunate that Adobe doesn't explicitly inform the user of this issue and I agree that it's time for an update in order to meet current user expectations.

[Denyerpro]

I don't suppose you chaps notice any performance drop / disk thrashing when attempting to perform simple operations at 100% zoom on large 16Bit images? On 36MP images, 16Bit, single layer, if I zoom to 100% and attempt to use for example a simple curves adjustment, not only is the screen update slow, but there is disk noise from the PC  -  Especially odd as none of the scratch disks, files, or applications are on mechanical drives in the system.