Interpreting histograms to predict banding in prints

[Jeffrey Saldinger]

The attached screenshot shows the histogram of an 8-bit grayscale PSD (it was made from the histogram in Photoshop). With those spikes in the histogram, wouldn’t one expect to see some banding, even on a relatively small 6.3x4.2 inch print?

I was relieved not to see any banding and have been curious about why I did not. Was it because there weren’t many areas of the sort that, as I understand it, are most susceptible to banding (sky, or areas of flat tone, or areas with smooth gradients without any details)? Or because the print was relatively small (and there might be some banding I’d see if the files were printed larger)? Perhaps something else?

It might be useful for me to add (and thus answer a question that may be asked) that I need 8-bit grayscale PSDs to send out to a printer with whom I’m working on a project for which the prints will be 6.3x4.2 inches. So I’m printing the files at home to see how they look before I send them to the printer.

Incidentally, in Lightroom the histogram for this file is an “ordinary” smooth one, very different from Photoshop’s. Why are they so different?

Thanks for your help.

[digitaldog]

In a word, no.

More about Histograms to help:

Everything you thought you wanted to know about Histograms
Another exhaustive 40 minute video examining:

What are histograms. In Photoshop, ACR, Lightroom.
Histograms: clipping color and tones, color spaces and color gamut.
Histogram and Photoshop’s Level’s command.
Histograms don’t tell us our images are good (examples).
Misconceptions about histograms. How they lie.
Histograms and Expose To The Right (ETTR).
Are histograms useful and if so, how?

Low rez (YouTube): http://www.youtube.com/watch?v=EjPsP4HhHhE
High rez: http://digitaldog.net/files/Histogram_Video.mov

[Jeffrey Saldinger]

Thank you, Andrew.

I'll watch the video and get in touch as might be appropriate once I've digested it

[Ernst Dinkla]

Thank you, Andrew.

I'll watch the video and get in touch as might be appropriate once I've digested it

And if the spikes in that histogram really result in banding then adding a tiny bit of noise could help.

Met vriendelijke groet, Ernst Dinkla

        T
SOLI  AIR
        D

"For every complex problem there is an answer that is clear, simple, and wrong." H. L. Mencken

[bjanes]

The attached screenshot shows the histogram of an 8-bit grayscale PSD (it was made from the histogram in Photoshop). With those spikes in the histogram, wouldn’t one expect to see some banding, even on a relatively small 6.3x4.2 inch print?

I was relieved not to see any banding and have been curious about why I did not. Was it because there weren’t many areas of the sort that, as I understand it, are most susceptible to banding (sky, or areas of flat tone, or areas with smooth gradients without any details)? Or because the print was relatively small (and there might be some banding I’d see if the files were printed larger)? Perhaps something else?

It might be useful for me to add (and thus answer a question that may be asked) that I need 8-bit grayscale PSDs to send out to a printer with whom I’m working on a project for which the prints will be 6.3x4.2 inches. So I’m printing the files at home to see how they look before I send them to the printer.

Incidentally, in Lightroom the histogram for this file is an “ordinary” smooth one, very different from Photoshop’s. Why are they so different?

Thanks for your help.

The combing in the 8 bit file histogram is the result of editing the original image where some pixel values are truncated. This could be eliminated or markedly reduced if you edit the file in 16 bit TIFF and reduce the final image to 8 bits.

The appearance of the histogram is affected by binning, where the x-axis values do not represent 1 pixel but a range of pixels. This is shown in the two raw histograms of Nikon D850 files shown below. With a bin width of 1/48 EV combing is prominent, while with a bin width of 1/12 EV combing is markedly reduced.

Theoretically, a raw file should not be pixelated, but the blue and red channels show more prominent pixelation due to the white balance pre-conditioning that Nikon does with its raw files. In the deep shadows, combing is increased because few photons are collected and some bins contain no values.

Bill

[Jeffrey Saldinger]

Andrew, thank you for referring me to your video. I’m glad to have been introduced to what you cover in it, especially (as it relates to my immediate concerns) what you address beginning around the 25 minute mark, where you begin to discuss the histogram that indicates the data loss from a jpeg after it was adjusted in Photoshop and then you show the same image similarly adjusted in Lightroom, where the image looks better and there is no combing of the histogram.

I’m still wondering about whether data loss revealed in a histogram can be used to anticipate banding in a print. Any further thoughts on this? Does one just have to make the print to see? Or can one count on seeing banding on the screen if there will be banding on the print? Perhaps there are too many specifics for a given file, and no generalities are useful here?

[Jeffrey Saldinger]

Bill and Ernst, thank you for your specific pointers.

[Jim Kasson]

Andrew, thank you for referring me to your video. I’m glad to have been introduced to what you cover in it, especially (as it relates to my immediate concerns) what you address beginning around the 25 minute mark, where you begin to discuss the histogram that indicates the data loss from a jpeg after it was adjusted in Photoshop and then you show the same image similarly adjusted in Lightroom, where the image looks better and there is no combing of the histogram.

I’m still wondering about whether data loss revealed in a histogram can be used to anticipate banding in a print. Any further thoughts on this? Does one just have to make the print to see? Or can one count on seeing banding on the screen if there will be banding on the print? Perhaps there are too many specifics for a given file, and no generalities are useful here?

With 8-bit precision images, you can have mild histogram depopulation with no visible print posterization.  You can also have visible posterization with no depopulation, particularly in CIELab.

Why not process in 16-bit (well, 15+) precision and put in a bit of noise if your source image is 8-bit?

Jim

[Jeffrey Saldinger]

Thank you, Jim. I had left out of the OP the backstory of why I'm working with 8-bit PSDs (instead of "15+"), but had written it up for another purpose. This is it (everything began with NEFs or DNGs and "all their bits"):

For a custom printing job of black and white photographs, I’ve been asked to submit to the printer flattened 8-bit grayscale PSDs at 300 ppi; they will be printed at 6.3x4.2 inches. Most of the images have no or little sky, or areas of flat tone, or areas with smooth gradients without any details.

Some of the images required going through as many as four proofs, and for each file that needed to be reworked (that is, for the 2nd, 3rd, and 4th proofs, as may have been necessary), I copied the previous 8-bit grayscale file and added to the copy the adjustments I thought would work. (I am aware that I could have gone back to the original PSD and reworked it rather than work on a copy of the previous small 8-bit grayscale file, but it was more convenient to do it the way I did).

For all the 8-bit files I sent for the first proof, the histogram was smooth; the data loss began to appear in those files that needed subsequent alterations.

[digitaldog]

I’m still wondering about whether data loss revealed in a histogram can be used to anticipate banding in a print. Any further thoughts on this? Does one just have to make the print to see?

Far too many variables just in the print output process to answer so yeah, you'd have to run tests. This will depend on the printing technology, the degree of sharpening for output, the paper surface, the image content (got synthetic gradients? Smooth skies?) the color space.
As Jim states, just do all the work high bit and its all moot. THEN save off the 8-bit data for the printer.

[David Sutton]

To add to what has already been said, a quick way to check for banding is to create a temporary solar curve on top of your layer stack (see attached photo). If you set it up as an action you only need to make it once.
If banding or other artefacts show up, progressively turn off the underlying layers until you find the culprit.
You'll quickly learn what will be visible in your final print.

Edit: It'll be easier if I just give you the .atn file here
I'll leave it on the website for a few weeks.
Half the time the banding I get is in the masks, YMMV.

[BobShaw]

Edit: It'll be easier if I just give you the .atn file here

Wow thanks. i was aware of the solar curve but never thought of an action for it. Ta.

[Jeffrey Saldinger]

Thank you, everyone. This has been gratifyingly informative and helpful.