Pages: [1]   Go Down

Author Topic: Structure and Clarity: what they really do, explained  (Read 797 times)

kpz

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 63
Structure and Clarity: what they really do, explained
« on: August 11, 2019, 03:43:05 am »

There are many threads on various internet forums asking about the details of these sliders, but no good answers. In this post, I endeavor to change that.

Structure is the usual "unsharp mask" sharpening used on every image editing program in existence. The twist is that the radius used scales with the size of the image. Large images mean bigger radii. This YouTube video contains proof: https://www.youtube.com/watch?v=P8jNw-y_CeU.

Clarity is a bit trickier. Let's start with the "Neutral" option with positive values. Here, C1 performs a global adjustment of the luma curve. The shape of the curve is a shallow S which crosses the input = output line (45 degree line from lower left to upper left corner of the curves window) at an L value somewhere in the low 30s, where I measure in L in the LAB color space. (Note the luma curve window uses different units.) In particular, neutral does not touch any color information. The A and B values in LAB stay the same. You can prove this for yourself by going to DPReview, downloading one of their test images with a Kodak grey step wedge, and trying to match the results of a positive clarity adjustment with a luma curve.

With "natural" and "punch," there are adjustments to the A and B values to make the colors more saturated. "Punch" saturates more than "natural." I did not check, but my (totally ignorant) guess is that they're curving in the A and B spaces too, so you could plausibly recreate these sliders in another editing program (e.g. Photoshop) just by figuring out the right curves from a synthetic test image of many color patches.

A point that surprised me is that clarity (at least neutral with positive values) is a totally global adjustment. The Capture One documentation makes it sound like local contrast enhancement. It's not!

I don't know what negative Clarity values do, but it seems like a different algorithm. The "neutral" option still operates totally on the L channel, but there are now local effects that can be detected by measuring LAB values on the transition regions between color patches. I think. I can't actually see them, and there's noise in the LAB measurements, but it replicated enough with different patches that I'm pretty sure of it.

I hope this was helpful, or at least interesting.
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8295
Re: Structure and Clarity: what they really do, explained
« Reply #1 on: August 11, 2019, 09:04:05 am »

There are many threads on various internet forums asking about the details of these sliders, but no good answers. In this post, I endeavor to change that.

Hi,

It's an interesting endeavor, if not only to create a better understanding of the behavior of these controls.

Personally, I'm a supporter of the alternative postprocessing tools offered by Topaz Precision Contrast (Clarity), and  Topaz Precision Detail. They address the same effects, but in a more detailed and controllable way. But if something similar can already be achieved during Raw conversion (and thus baked into the adjustments saved with the particular variant), that would make life easier (and still leave the option of switching it off, and use postprocessing).
 
Quote
Structure is the usual "unsharp mask" sharpening used on every image editing program in existence. The twist is that the radius used scales with the size of the image. Large images mean bigger radii. This YouTube video contains proof: https://www.youtube.com/watch?v=P8jNw-y_CeU.

I'm not so sure that 'sharpening' is the mechanism used, it seems more like a Wavelet transform, with certain high frequencies boosted. The video only looked at a simple high contrast edge transition, and may therefore reach a wrong conclusion. A wavelet boost can also result in overshoot and undershoot at sharp edges, if overdone. It might be worthwhile to apply it on a stepwedge, and see if it behaves the same at different luminosity levels, and it may help to first apply a little blur and see how much of that can be recovered by contrast /amplitude manipulation instead of Deconvolution.

Quote
Clarity is a bit trickier. Let's start with the "Neutral" option with positive values. Here, C1 performs a global adjustment of the luma curve. The shape of the curve is a shallow S which crosses the input = output line (45 degree line from lower left to upper left corner of the curves window) at an L value somewhere in the low 30s, where I measure in L in the LAB color space. (Note the luma curve window uses different units.) In particular, neutral does not touch any color information. The A and B values in LAB stay the same. You can prove this for yourself by going to DPReview, downloading one of their test images with a Kodak grey step wedge, and trying to match the results of a positive clarity adjustment with a luma curve.

I'm not sure how Clarity is applied in Capture One.bClarity is usually explained as a medium-luminosity local contrast boost, or something like a HiRaLoAm (High Radius Low Amount) USM operation. I'm not sure if the Capture One implementation is really that simple, especially given the different flavors that it offers.

Clarity is implemented quite differently in Topaz Precision Contrast, and I like that type and level of control very much. It can give an almost 3-dimensional look to images. But as said, a better understanding of how Clarity works in Capture On is welcome.

I think both controls need more investigation to really understand what is going on under the hood, but this is a helpful start.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

kpz

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 63
Re: Structure and Clarity: what they really do, explained
« Reply #2 on: August 11, 2019, 09:37:10 am »

I'm not so sure that 'sharpening' is the mechanism used, it seems more like a Wavelet transform, with certain high frequencies boosted. The video only looked at a simple high contrast edge transition, and may therefore reach a wrong conclusion. A wavelet boost can also result in overshoot and undershoot at sharp edges, if overdone. It might be worthwhile to apply it on a stepwedge, and see if it behaves the same at different luminosity levels, and it may help to first apply a little blur and see how much of that can be recovered by contrast /amplitude manipulation instead of Deconvolution.

Please see the attached image, which is an edited test chart raw file from DPReview. The camera used is a Fuji G50R at 100 ISO, and I turned "structure" to 100 and kept everything else on the defaults. It sure looks like a low-radius USM to me.

Quote
I'm not sure how Clarity is applied in Capture One.bClarity is usually explained as a medium-luminosity local contrast boost [...] I'm not sure if the Capture One implementation is really that simple, especially given the different flavors that it offers.

For positive values on the "neutral" setting, I assure you it is indeed that simple. It really is just a medium-luminosity boosting global curve. If you are skeptical, I invite you to replicate the results with a step wedge.

Quote
Personally, I'm a supporter of the alternative postprocessing tools offered by Topaz Precision Contrast (Clarity), and  Topaz Precision Detail. They address the same effects, but in a more detailed and controllable way.

Any guesses what the two Topaz alternatives do under the hood? It looks like Clarity is wavelet-based. I'm not sure about detail.
Logged
Pages: [1]   Go Up