How do you use Capture Sharpening?

Hello everyone, I ask you, how, you use the new Capture Sharpening option present from v.5.8. Do you use only that or also associated with the sharpening panel options? And in the case with which parameters?
Thank you

The feature is so new that we were hoping you would tell us!

@carlos1960 Although the automatic features are pretty good, I like to keep things manual and more moderate:

  • Contrast threshold = 0 (reason: the entire image has been subjected to the ‘same’ blurring, so we should perform the same deconvolution everywhere)
  • Radius = 0.5 (reason: higher values seem to do too much for my liking and may introduce artifacts)
  • Corner radius boos = 0 (reason: haven’t really seen the benefit yet, but I might be convinced otherwise)
  • Iterations = 20 + Auto limit iterations (reason: this works really well in the majority of cases)
2 Likes

I’m experimenting with it but I’m very undecided whether to use Capture Sharpening and Deconvolution together, or just Capture Sharpening.

…thanks for the informations. But do you only use Capture Sharpening or also something from the sharpening panel?

My current workflow is changing a little bit, but I am usually happy with just CS and then a little sharpening on resize.

1 Like

Hi @carlos1960

You can find a draft copy of the documentation here:

I will move this to RawPedia soon.

I use capture sharpening on automatic settings, no normal sharpening, and always post-resize sharpening.

2 Likes

As almost anything related with photography, there’s not a single, true answer to your question. It will depend on a few factors: your sensitivity to noise, if you will be downscaling the final image, the quality of the input raw image, the amount of processing you will be doing to that image, …

I have a tendency to think about Capture Sharpening (CS) as a tool to take the image to the best starting point for other tools. To me, it’s just as important as the correct demosaicing algorithm, the white balance, and choosing the right color profiles.

If you have a really nice sensor, with an outstanding lens, coupled with a perfect shooting technique, then you will need little to no sharpening applied to your images. On the contrary, if you fail on any of those points, then you will need to enhance your images.

In other apps workflows sometimes people talk about the need of pre-sharpening the images. There’s a lot of debate about that. In RT, with CS you will pre-sharpen your image. How much will depend on how much you will further develop it: the more you play with contrast (tone mapping, contrast by detail levels, wavelets, local contrast, …), the more you will see the artifacts introduced by CS with strong settings. In my workflow I tend to be gentle with CS settings, as it won’t be the only tool used.

As Thanatomanic said, most of the times I find myself lowering the radius and iterations from the auto-calculated values. They tend to give halos not really visible at 100% or lower zooms, but not desirable in my workflow (for focus stacking). Radius between 0.46 and 0.60 are usual. And iterations between 5 and 15 are usual, too. As you may guess, with those values the sharpening is not so WOW!, but it helps the other tools to better do their jobs.

On the contrary as Thanatomanic, I usually leave the auto-contrast threshold value as is, or maybe increase it a bit. The idea is to not sharpening noise, as in focus stacking it has a tendency to increase exponentially.

If your image will be downscaled…, well…, you absolutely need a post-resize sharpening, as almost any sharpness improvement you get with CS will be lost. The aid of CS to other tools of the processing will be there, yet. So even if I downscale my images, to me CS is a must, and should be almost always on.

As it is said somewhere, CS works better with the unsharp mask sharpening. In my experience I have used it with both US mask and RL deconvolution:

  • with unsharp mask: I tend to use this method with a Hiraloam approach, that is, looking to enhance the perceived volume of the subject, as I have already played with contrast during the developing process, and the image already looks sharpened
  • with RL deconvolution: in this case the image lacks a bit of crispness, so I use this method with a high threshold to enhance the most prominent details, while leaving alone the tiniest details (and remaining noise). Note in this case that this tool has a tendency to quickly generate halos

Hope this helps. But honestly, the better way to find your best workflow is by playing with your images.

5 Likes

Ha, I am a bit lax with CS and rely on the auto controls. Typically, if there wasn’t auto, I would be more conservative, using smaller values. As for boost, I check reputable lens reviews to see whether I need to turn it on and to which degree.

1 Like

I also thank you for your explanations, I almost exclusively make photos of architecture and interiors for hotels for which I always work with tripod and low ISOs.

You’re welcome.

Then you may wish to test different settings and tools with your own shots.

Start with auto-calculated values in CS, then sharpen with unsharp mask to get a result you (or your customers) like. That won’t always mean the results with the least artifacts (halos).

Then try RL deconvolution with moderate settings, and choose the best approach.

1 Like

I absolutely agree with @XavAL :+1: since I follow the same path.
Thank you for your well-written contribution.
As I started with RAW-processing in 2013 I already realized that there shouldn’t be a “global” processing setting for many (different) photos - rather each individual photo should have its own workflow with unique adjustments - depending on the quality of the source material and the desired result / output.
CS is imo a very good tool to get a solid base for other (sharpening) adjustments, if necessary.
Most of the time I already get good results with the default settings (automatically). I rarely had to increase the radius a little bit, but nothing else.
And since I work with Sigma’s ART lenses, there’s often no need for additional sharpening…
In most cases, I only use capture-sharpening together with post-resize-sharpening (which depends on display- or printer output and the final dimensions of the photo of course).
So thanks again to all the CS developers for deleting another intermediate step from my (RAW-)photo workflow!
Regards :v:

4 Likes

One thing to keep in mind: If your camera has an antialiasing filter (basically a low-pass filter) in front of the sensor, then some sharpening may be needed even if the rest of the system is perfect.

2 Likes

Some sharpening is always required at various points of the pipeline. Ideally, it would be catered to the lens’, camera’s and scene’s physical properties and settings, and the rest of the entire workflow but we tend to use general tools because efficient and timely per image per step blur estimation and correction is still impractical. CS is a step in the right direction. That is why I have been giving feedback and encouragement from the start.

There might be a need for some tweaking of the auto controls but as can be seen above some people prefer lower settings and other higher.

2 Likes

I mostly use CS at default settings.

Exceptions are:

  1. Images with really long exposure time where hot pixles occur. In that case I enable raw hot pixels filter, which automatically recalculates the CS radius

  2. Images shot at high ISO where CS auto-calculation of Contrast Threshold (zero) and radius (too low) fails

4 Likes

I wonder how much CS can factor in noise and hot pixels. I sometimes have trouble getting rid of the latter using the pre processing filter.

For hot pixles try http://rawtherapee.com/shared/test_images/renon_earth_pyramids.pef

Are you asking me to verify? The statement above refers to some Play Raw files I encountered. I couldn’t get rid of the hot pixels, perhaps because they were in a large cluster in one of the cases. I dealt with them in G’MIC after export but by then there were already blurred bigger due to demosaicing, etc.

I have tried Capture Sharpening in the official 5.8 release, and I can say that I love it, as the linear colorspace contrast masking seems to make it easier to target the noise threshold for all of the tones in the image. When using old RL deconvolution, I had to choose between using a low threshold, perfect for the highlights with low amplitude fine detail above the noise threshold, but sharpening lots of shadow noise, or high threshold, not sharpening shadow noise, but getting waxy undersharpened highlight detail. CS contrast threshold just seems to work better relative to the natural noise w.r.t. tonal value curve of digital cameras. Also, negative halos and negative deviating noise amplification tends to be less objectionable than positive halos and hot pixel amplification.

As far as use, I generally set the radius manually to be as large as possible without creating halos in the most in focus region in the image. Corner radius boost, I do not find particularly useful as usually the corner unsharpness tends to be greater in either the tangential or radial direction, rather than a uniformly larger blur, so corner radius boost results in one axis being insufficiently sharpened and another axis being sharpened to the point of haloing. I either set the contrast threshold to zero, if I want to have a grainy textured image, or if I want clean, I manually set to be as low as possible without sharpening noise. Iterations, I use lots for low ISO, and less for high ISO images, just like I used to for RL deconvolution. For other sharpening, I find CS results in improved resolution, but not crispness, so I add unsharp mask at a very low amount on top of CS, and find the two to be nicely complimentary.

With high ISO, I use less capture sharpening iterations, no unsharp mask, and a little bit of Edges sharpening so as not to amplify noise as much.

3 Likes

I just enable it and smile!

3 Likes