Radius used in USM sharpening

[relic]

In another thread, @pdk42 stated:

I found that PP sharpening was problematical on HR photos using LR. The best I got was to increase the Radius quite a bit. Just cranking up the Amount on its own was not enough. I suspect that other raw tools might provide a better result than LR though. Try Oly's Workplace - see what that does.

I may be totally wrong, but I've been in the habit of changing the radius in USM according to the resolution of the image, my thinking being that the radius should correspond to the size of the "image unit" (whatever that means). So for images downsized to 1200 or so pixels, I use a radius of 0.3, whereas for images at about 3000 - 4000 pixels, I use a radius of 0.5 to 0.7 pixels; so if my approach is right (and that is questionable in my mind) then HR images would require even larger radii, which is in accord with @pdk42's experience. I would appreciate comments on whether I am wrong. Many thanks.

 

[pdk42]

My own experience is that a value for radius of about 2 works the best.

 

[relic]

My own experience is that a value for radius of about 2 works the best.

Thank you. My point, though, was that I find that for a 1000 to 1200 pixel size image, a radius of 0.3 works well, but has very little effect when the image size is 4000 pixels, where, for the same effect, I need to double or triple the radius. So, to me, the fact that you need to increase the radius for a higher resolution image is not surprising. But again, my perception may be erroneous.

 

[ArizonaMike]

Personally I have also found that the radius has a huge impact on sharpening of HR images, but I think it is worth keeping In mind that various tools act differently. That is, that the meaning of a radius of 2 is different for LR than for Oly Workspace or CaptureOne or Dxo's PhotoLab or ... So perhaps a specific value for a radius for one USM in one tool is meaningless for any other tool.

 

[mfturner]

I agree that different tools must implement USM differently. And your own taste for sharpening comes into play, as well as whether you zoom into the image. If you zoom in with the same magnification, you will likely be happier with the same USM radius regardless of resolution. However, if you don't tend to zoom in (or with prints if you back up to view larger prints as was the old assumption), then you will likely be happier scaling the sharpening radius with picture width or height.

I also often scale my sharpening radius if I don't expect people to spend the time to zoom in. My exception was when I was taking photos of the high school band or choir, because I know parents zoom in to see their own kids faces.

 

[Machi]

In another thread, @pdk42 stated:

I may be totally wrong, but I've been in the habit of changing the radius in USM according to the resolution of the image, my thinking being that the radius should correspond to the size of the "image unit" (whatever that means). So for images downsized to 1200 or so pixels, I use a radius of 0.3, whereas for images at about 3000 - 4000 pixels, I use a radius of 0.5 to 0.7 pixels; so if my approach is right (and that is questionable in my mind) then HR images would require even larger radii, which is in accord with @pdk42's experience. I would appreciate comments on whether I am wrong. Many thanks.

Sharpening radius depends mostly on three things.
1) How good deBayer process is (more important for sharp lenses).
2) How large are smallest details in the image.
3) Which kind of sharpening one uses.

Generally principle is:
- more blurry image = larger details visible -> larger sharpening radius.
- less blurry image = smaller details visible -> smaller sharpening radius.
So it's not primarily dependent on size of image as even full size image can show small details with good sharp lens. Only thing which resizing could do is that if original image has relatively large details which are small after resizing then one needs to use smaller radius.

In Rawtherapee, which has one of the best deBayer algorithms, I use mostly radius 0.4-0.8 for normal photos sharpened by deconvolution and 1.2 for hires photos taken under optimal conditions.
Only when I miss focus or is my image blurred by atmospheric conditions then I use larger radius.

 

[relic]

Sharpening radius depends mostly on three things.
1) How good deBayer process is (more important for sharp lenses).
2) How large are smallest details in the image.
3) Which kind of sharpening one uses.

Generally principle is:
- more blurry image = larger details visible -> larger sharpening radius.
- less blurry image = smaller details visible -> smaller sharpening radius.
So it's not primarily dependent on size of image as even full size image can show small details with good sharp lens. Only thing which resizing could do is that if original image has relatively large details which are small after resizing then one needs to use smaller radius.

In Rawtherapee, which has one of the best deBayer algorithms, I use mostly radius 0.4-0.8 for normal photos sharpened by deconvolution and 1.2 for hires photos taken under optimal conditions.
Only when I miss focus or is my image blurred by atmospheric conditions then I use larger radius.

Thank you. I agree. That is what I meant by the term "image element" in my original post (I didn't know what to call it-- I guess I could have called it the size of the finest detail or something like that). And I should have clarified that when I said it depended on image size I was thinking the same image sharpened at different pixel sizes-- apologies.
Thanks to all for clarifying that in different program the radius size means different things: I was not aware of that.

 

[ArizonaMike]

It would be good to have some idea of the relative quality of USM sharpening for different photo apps. That is, which photo editor should someone use if they want to make sure they get the best sharpening that they can.

Based on the comment by @Machi I spent some time working with RawTherapee which I had downloaded some time ago but largely ignored to see how it sharpened by High Res photos which always seemed a bit soft to me, regardless of the lens and aperture I used, and found that it did a wonderful job - better than either of the normal workflow tools I had been using.

I just don't think there is any independent (and thus objective) way of comparing the way different apps handle sharpening.

 

[Hoffelijk]

About USM sharpening I found this article at Cambridgeincolour educational.

Guide to image sharpening

 

[ArizonaMike]

About USM sharpening I found this article at Cambridgeincolour educational.

Thanks for posting that. I found out the actual definitions for some of the sharpening terminology so it was very helpful.

As I mentioned in an earlier post I started playing with RawTherapee and, for once, actually started reading the documentation and one of the things it said was to not do any image rotation until the end of processing beause doing it early in processing impacts sharpening. That is, edges will be distorted by image rotation and that means that fixing them may require more sharpening later in the processing than would otherwise be necessary and result in an over sharpened image.

Never thought about that until I read the manual. I wonder what other good advice I will come across when reading through the manual.

 

[Hoffelijk]

Indeed as far as I have always understood, first work on the original without sharpening, do your favorite adjustments and only when you are at the point of the end result (webpublish, resolution or print) you start exporting sharpening.

In Workspace with the highest 16MB resolution of my E-PL8 I do have a default USM value to start with.

Strenght: 220
Radius: 0.4
Threshold: 1

 

[relic]

Indeed as far as I have always understood, first work on the original without sharpening, do your favorite adjustments and only when you are at the point of the end result (webpublish, resolution or print) you start exporting sharpening.

In Workspace with the highest 16MB resolution of my E-PL8 I do have a default USM value to start with.

Strenght: 0.4
Radius: 220
Threshold: 1

Just wondering: did you have the Radius and Strength values accidentally reversed?

 

[Hoffelijk]

Just wondering: did you have the Radius and Strength values accidentally reversed?

Jup

corrected that in the post.