USM blending options vs. decompose / recompose

GIMP help for unsharp mask describes decomposing an image to HSV, sharpening value only then recomposing (the same thing can be done via Lab / L). There’s no explicit “sharpen luminosity only” setting in the USM plugin, but there is the Blending options list which includes HSV Value, LCh Lightness and Luminance modes.

Do they achieve the same thing as the decompose / recompose process?

I also tend to use repeated very light USM passes (R = 0.5, A = 0.75, T = 0, tweaked as needed) rather than one heavy pass. It seems to look better and I can walk back-and-forth through the history stack to find the best spot.

Any recommendations?


I also noticed a blend option “Replace” in the filter dialog. When the Reset button is pressed, Replace is selected, but I’ve not (yet?) found it documented.

Any clarification?


Likely. You could test your hypothesis. :wink:

Single pass is cleaner. Iteration stacks existing and USM artifacts. Is there a USM tool window with a preview toggle? If so, that would be better than using history.

I haven’t used GIMP in ages. That seems to indicate the blend replaces the layer being filtered with the filtered layer. Just a guess… Again, experiment. :wink:

I did sharpen a couple of images and didn’t notice any real difference between the two approaches, but image ops can be (naturally) highly dependent on the image and other factors. So I was kinda hoping to get a more objective confirmation from someone who possibly “knows the code” rather than based on my empirical testing. :slight_smile:

As to USM, years ago (mid 2000s) I remember reading Jeff Schewe (maybe?) and others – probably on Luminous Landscape – say multiple light passes were preferable to fewer or even a single heavy application. But that may have changed. By using the history tool in GIMP, I can visually compare the changes to the image as I walk up and down the history stack.

Actually, changing the blend option in a filter apparently applies the changes made by the filter directly to the target layer (with the chosen opacity), so there’s no need to work on a duplicate layer,* choose blend / opacity there and then merge that layer with the original to achieve the same result. Effectively, the previously post-filter blend / opacity / merge ops have been moved into the filter. At least that’s how it seems to work.

* Unless a dupe is made as a backup.


I will leave the Gimp GEGL unsharp mask to you clever guys

…but…the Gimp gmic plugin does have an unsharp mask filter with ‘gaussian’ and HSL(lightness) options.

Obviously Gimp settings USM passes (R = 0.5, A = 0.75, T = 0 are not applicable, but you might get something out of it.

Yeah, I’ve got G’MIC installed but hadn’t looked there yet. Guess I need to find their documentation and start reading…


Iteration is not effective. You may try cumulative vs single. Start with the former, then find a setting that comes close with the latter. The latter will be cleaner when observed independently of other operations such as scaling, interpolation and compression. USM is a crude way to sharpen.

People often use it because it is easier to understand and get quick results. However, filtering has move on to context, noise and structure aware methods. A comparative method research papers use is how many iterations can a filter tolerate before degradation happens. USM is baseline.

USM still exists but usually used at the beginning and/or end of the pipe to limit its adverse impact on the image being processed: the so-called “capture” and “post-resize” sharpening.

That is what I was guessing. Sorry, if I wasn’t being clear. In-place filters are useful in that we don’t have to worry about future layer management and flattening. Having a preview option makes the alternative “create result in new layer” irrelevant.

PS That said, there is one approach that would warrant multiple passes: if you were targeting different frequencies (i.e. radii) and using different amounts and thresholds for each.

PPS There is also a possibility GIMP is using a modified USM that produces cleaner results.

I tend to probably under-sharpen to avoid artifacts as much as possible. But I’d like to get sharper, cleaner results. At this point I’m interested in best practice and tool recommendations (which indirectly drove my original question).

I’m currently using a combination of darktable (mostly, sometimes RawTherapee) to process and GIMP to finalize the images. I am far from expert on either. That’s not to say I might not try other tools at some point (free or otherwise) but it’s where I am now. I’m also currently on Windows after years of Linux, which could also influence my choice of tools.

Of course, my first priority should be to start shooting images worthy of the effort! :slight_smile:

The key to success is to develop a good sense of workflow and how everything works toward the goal: e.g. sharper and cleaner. We often do things because someone says it is a good idea. They may not necessarily be wrong, but in our context, not correct either. In sum:

  1. Experiment to see what works.
  2. Change it up (e.g. different radii per iteration) to see if that improves your method.

Keep us posted on where your journey goes. Ask questions and I am sure the community will be glad to offer ideas (and opinions :smiling_imp:).

PS Specifically to your inquiry, I mentioned doing sharpening at the beginning or end of the pipe. dt also offers AP’s modules that offer sharpening. Another idea would be to remove or otherwise work on the artifacts before sharpening (e.g. simple Gaussian blur or advanced dt denoising). That way, you wouldn’t be sharpening noise or artifacts but instead enhancing the good parts.

Yeah, about 15 years ago I was shooting a bit more. Back then it was with an 8mp DSLR, CaptureOne and Paint Shop Pro. With that combo, USM worked pretty well, whether for sharpening or “LCE”, local contrast enhancement, as they called it then. Then I kind of fell out of shooting for a while and just started again in late 2019. Now my interest is more landscape focused than before.

I’ve usually denoised as early as possible (unless it’s needed again after a potentially noise-inducing edit) and then sharpened after final sizing, for the most part. As you indicated, I take care of issues before sharpening, such as darken-only-cloning to remove sky halos, etc.

1 Like