Camera Raw/Lightroom" Unpleasant behaviour of sharpening

  • 2
  • Problem
  • Updated 1 month ago
  • (Edited)
I've noticed that sharpening (from details panel) in Lightroom (on different platforms) as well as in Adobe Camera RAW produces very unpleasant high-contrast labyrinths (sample attached). This forces me to disable sharpening at all and switch to Photoshop where I can apply Unsharp Mask (which behaves normally in my opinion).

I don't know if it's a bug, but I'd like to have sharpening mode option similar to Unsharp Mask to be available in LR/ACR please (for example as additional switch in Advanced settings), because it is very annoying to be forced to switch to PS every time I need proper sharpening.
Photo of Rath5

Rath5

  • 12 Posts
  • 2 Reply Likes

Posted 2 months ago

  • 2
Photo of Rikk Flohr

Rikk Flohr, Official Rep

  • 7545 Posts
  • 1729 Reply Likes
At what % Zoom do you have set in this view?
Photo of Rath5

Rath5

  • 12 Posts
  • 2 Reply Likes
Please see my new images in the answer to E. Dinur below (click to enlarge in new window).
Especially look at bottom right part leaves. You can notice loss of contrast and well defined lines forming sort of mazes.
And this is with Detail 0/Radius 1.


Also look what happens with other combination (compared to disabled Sharpening):


(increased radius gave thicker maze lines, masking made some blurry areas in the center of leaves). Today also spotted same effects in Photoshop Express for Android.

Compare this to USM with radius 2.0 (no mazes, edges become well defined):



Something is wrong here, ACR Sharpening basically is not usable due to this artifact.

Original:


Photo of Doug S

Doug S

  • 48 Posts
  • 15 Reply Likes
Did you apply a mask?
Photo of Rath5

Rath5

  • 12 Posts
  • 2 Reply Likes
no mask
Photo of Dave Grainger

Dave Grainger

  • 442 Posts
  • 82 Reply Likes
I think that any image blown up / zoomed that far so that you are seeing individual pixels would look like that! It appears to me that you do indeed have a pretty sharp boundary between the two areas...
Photo of Rath5

Rath5

  • 12 Posts
  • 2 Reply Likes
no, it's about unusual behaviour of ACR sharpening which creates mazes, I'll post more images
Photo of E. Dinur

E. Dinur

  • 13 Posts
  • 4 Reply Likes
In Develop/Detail put the Sharpening Detail slider to 0 and you will have USM. At 100 it is Deconvolution sharpening and in between it is a variable mixture of the two. Increase Radius to soften tonal transitions.
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
Sorry, Dave, but you're the one who's not learning or listening here. The sharpening artifacts in ACR/LR are completely different than the artifacts created in either Ps or other raw processing software. The issues in ACR/LR have been around since day 1 and i flagged the issue more than 15 years ago. It's an artifact of their sharpening algo and it just looks bad.The only way to avoid it, is either by not using sharpening in ACR/LR or to add grain after sharpening to break up the pattern regularity which Sharpen creates. Most professional retouchers i know (100s) either use C1 for raw processing or will simply not sharpen in LR if that's what they're asked to process with.This is common industry practice and knowledge.
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
Here's an example of what's going on in ACR vs PS. I've maxed out both sharpen settings to illustrate the difference and have split the image in half... can you tell which is ACR?
Which of these looks most natural to you?
Photo of Dave Grainger

Dave Grainger

  • 442 Posts
  • 82 Reply Likes
The one on the right side looks better to me. Which is that one, ACR or PS?
Photo of Dave Grainger

Dave Grainger

  • 442 Posts
  • 82 Reply Likes
eartho: Thanks for the info. I had not looked at Capture One, so after your comment, I downloaded it and am going through the intro stuff. Also: even starting into my ninth decade next month, I am open to learning...  I don't use Lightroom, just Bridge and CC 2020, so was not familiar with the sharpening in LR.. Dave
(Edited)
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
Hey Dave, yes LR/ACR is on the left, with Ps on the right. You can also click on the image to see a larger version. 
Here's another comparison of the same image, but i've added the original as reference and the image is shown at 200%.
Click through to see full size.
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
Yes, i bugged this all the way back in 2006 with LR V1. LR/ACR has the worst sharpening algo of any processing software i've ever used... and i've tested it against ALL OF THEM.
The maze pattern is just awful and actually adds texture which doesn't exist in the natural grain. I seriously doubt that grain patterns are consistent at all ISO and across all sensors, yet with enough sharpening in LR, the same pattern emerges.
The only way i've found to reduce the pattern after sharpening is by adding grain back into the image to break up the maze regularity.
Photo of Rikk Flohr

Rikk Flohr, Official Rep

  • 7545 Posts
  • 1729 Reply Likes
Please provide the bug number and I will review it. 
Photo of Todd Shaner

Todd Shaner, Champion

  • 1961 Posts
  • 664 Reply Likes
The maze pattern is just awful and actually adds texture which doesn't exist in the natural grain.
eartho and Rath5 what will really help is to provide a raw file that exhibits the issue. Apply your LR/ACR Develop module settings  for "best results" that exhibit the issue and export the file to DNG file format. Post the file to Dropbox or other file sharing site and provide the share link in a reply here.

LR/ACR Sharpening is much more complex than RGB USM and uses algorithms designed for specific RGGB camera sensor technology (Bayer and Fuji X-Trans). The fact you are seeing USM differences in LR/ACR export files with no sharpening applied is expected and not recommended. In fact the results should be inferior to LR/ACR sharpening. I suspect there's something else that's causing what you're seeing. When I see what you call grain pattern it's usually easily fixed by adding some Luminace NR and then adjust sharpening as required.

Example raw files with your settings will help a lot here! Thank you!
Photo of Dave Grainger

Dave Grainger

  • 442 Posts
  • 82 Reply Likes
I just read through the entire thread: my head is spinning! (that 8 year old post)
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
Here's a test for you all... In LR/ACR, open a raw, zoom in 200%, load the Version 1 engine and set Sharpen to 150, 1.0, 100, 0. That's how (exaggerated) sharpening should look. Without changing any settings, go back to Calibration and load Version 2... Yuck!
I was mistaken earlier thinking that the sharpening was bad in v1. I must have been having a senior moment =)
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
and from that thread, this statement is not true:
>Detail -- Adjusts how much high-frequency information is sharpened in the image and how much the sharpening process emphasizes edges. Lower settings primarily sharpen edges to remove blurring. Higher values are useful for making the textures in the image more pronounced.

Higher values do not make the natural textures more pronounced... higher values introduce a false texture which doesn't exist in the original image.
Photo of Robert Somrak

Robert Somrak, Champion

  • 724 Posts
  • 227 Reply Likes
Process 1 vs 2 was pretty extreme!!!
(Edited)
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
Robert, yes! That's where it all went wrong. I have no idea what their intention is with the detail slider, but the texture it creates is very unnatural and should always be avoided. 
To test whether or not Detail uses deconvolution for its sharpening, i ran decon in Raw Therapee and even with everything maxed out, a maze pattern does not appear. There's simply something wrong with the algo for Detail.
Photo of Robert Somrak

Robert Somrak, Champion

  • 724 Posts
  • 227 Reply Likes
Rath5
Why do you think the 150 Amount in Lr is the same as 150 in Unsharp Mask?  That is the maximum in Lightroom.  Crank the Amount up to the maximum in Unsharp mask and it is even worse.  There may be issues with sharpening is Lr but your comparison test is meaningless.  
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
i totally disagree. In my 3-up sample above, the sharpening in Ps is cranked to the max of 500 and looks 10x more natural than the LR sharpen. Click the image to see the full res and tell me which you think looks the best.
Photo of Robert Somrak

Robert Somrak, Champion

  • 724 Posts
  • 227 Reply Likes
My few tests show it just as bad or worse.  All the tools will break if you push them too far on certain images and these are good examples of that.  
(Edited)
Photo of eartho

eartho, Champion

  • 1480 Posts
  • 497 Reply Likes
again, i disagree. Both Unsharp and Smart Sharpen, will produce acceptable results when pushed to their limits, but ACR sharpen produces totally unacceptable results long before maximum.
Photo of Todd Shaner

Todd Shaner, Champion

  • 1961 Posts
  • 664 Reply Likes
Looking at this objectively I dipped into LR's arsenal of tools to see if there is a way to match PS's USM rendering. It turns out the Basic panel 'Texture' slider coupled with some changes in the Sharpening panel not only do that, but with less haloing.
Try the following settings:
LR: Texture = 50, Sharpening Amount = 100, Radius = 1.0 Detail = 10
PS: Unsharp Mask Amount = 300%, Radius = 1.0, Threshold = 0
This is a higher than normal amount of sharpening, but demonstrates the ability to "match" PS USM rendering in LR. I tested Canon 5D MKII CR2 files with a couple of subjects. Different camera models and subjects may require slightly different settings. The key takeaway is to 1) use a low Sharpening Detail setting and 2) apply the Basic panel Texture control. Give it a try and let me know your results.
(Edited)
Photo of Robert Somrak

Robert Somrak, Champion

  • 724 Posts
  • 227 Reply Likes
I don't know what photos Adobe tested the Detail slider with but I have yet to see a photo that looks better using HIGH detail no matter what the other sliders are.
Photo of Rath5

Rath5

  • 12 Posts
  • 2 Reply Likes
Decreasing Detail slider value or even moving it to 0 is not a cure from mazes.
Playing with Texture slider also did not help. I would not call this a match (USM looks heavier if not talk about mazes)

Photo of Robert Somrak

Robert Somrak, Champion

  • 724 Posts
  • 227 Reply Likes
BTW,  I still haven't figured out what the Alt/Option Detail slider view is supposed to be showing.  It seems to have no relation to the effect of the Detail slider.  The change of the Alt/Option Detail view can look minimal but the true effect of the Detail slider is a LOT more.
Photo of Todd Shaner

Todd Shaner, Champion

  • 1961 Posts
  • 664 Reply Likes
Bob, try 2:1 Zoom view. On my system the ALT/OPT key Detail control view is very representative of what it's doing.
Photo of Robert Somrak

Robert Somrak, Champion

  • 724 Posts
  • 227 Reply Likes
Todd,

Thanks, 2:1 Zoom is closer to what is happening.  Detail only works at higher setting on a super clean image and those kind of images really don't need it.
(Edited)
Photo of Todd Shaner

Todd Shaner, Champion

  • 1961 Posts
  • 664 Reply Likes
I generally use Detail = 35 for my Canon EOS raw files and only lower it for very noisy high ISO files that don't respond well to the Luminance NR control.

Bob, give my above suggestion a try concerning using a low Detail setting and the Texture control to emulate USM sharpening. It's a way to squeeze out some additional sharpening without increasing noise or creating artifacts.
Photo of Robert Somrak

Robert Somrak, Champion

  • 724 Posts
  • 227 Reply Likes
Thanks for the info Todd,  I will try it
Photo of Rath5

Rath5

  • 12 Posts
  • 2 Reply Likes
Also reposting my warning here: no combination of Detail and other sliders values can prevent described artifacts, artifacts can be only less or stronger but they are always present.
Photo of Jerry Syder

Jerry Syder

  • 499 Posts
  • 227 Reply Likes
I think your message is getting lost in this long thread - Adobe would surely be reading this post and take appropriate action where needed.