Upload & Sell: Off
First it helps to understand how USM works psychologically and technically. Psychologically sharpening is an optical illusion. Technically it is created by detecting borders between contrasting tones/colors and increasing the contrast at the border. By way of analogy it's like creating a stone wall along a property line. The three controls in the basic USM function control the characteristic of the "wall":
Amount = Amplitude (how much contrast is changed)
Radius = Width (how far on each side the contrast is applied)
Threshold = When to apply it (How much difference there needs to be between tones).
This is a screen shot of a tone border enlarged 800% in Photoshop created by hand (not in a photo):
Here it is with (500, .2, 0) USM applied:
As you can see USM alters the contrast at the borders. If you get up and walk across the room the eyes will not resolve the differences in contrast but it should seem sharper.
Some degree of USM improves nearly any photo because the capture process uses an RGB mosiac and anti-alias filters over the sensor are used to prevent that matrix from creating a "stair-step" pattern along diagonals. There's also the variable of lens resolution between lenses and different degrees of resolution at various f/stops that make USM desireable but the amount needed something which can only be determined visually. Even when a resolution chart is shot in a test it the eyes and brain which tell us which looks identical to how the eyes resolve the same chart directly.
Since images used on screens vary in size in a limited range and are viewed at reading distance it's pretty easy to arrive at USM settings you personally find work best by trail and error. Printing adds additonal variables: mechanical variables of the printing process, how the image is resized, and how far away it is seen. The same photo viewed on screen as a thumbnail, part of a web page, as screen wall paper often will benefit from three different levels of USM to look sharpest psycholgically when seen.
Back in '94 when I bought my first .8MP digital camera, an AppleQuicktake100 the problem was compensating for the loss of resolution when up sampling the small camera files for both screen and print. Printed at 300 ppi the out of camera file produced an acceptably sharp print, but it was the size of a postage stamp.
Today there is the opposite problem. The camera files are so large (pixel dimensions) reducing the pixel dimensions of the camera to screen size requires averaging them together. The greater the reduction in pixel dimensions from camera > screen display the more the downsizing will affect resolution of file details and the more USM is needed to enhance the contrast at the increasingly blurred tonal boundaries. If you take the same file of a forest scene and resize it three ways for screen -- thumbnail, 600 x 900 and full monitor size you'll likely find you need the most USM on the thumbnail because the downsizing turns the trees into green blobs. Viewed on the "pixel peeping" level the image will look bad, even to the point of clipping small highlights, but at normal viewing distance is will better create a sense of 3D texture.
USM is not a new trick created for digital. I first used it on camera back color separations in the 70's where a unsharp continuous tone image was created and sandwiched between the RGB images and screen when making the halftone color separation. Drum scanners created the same effect of adding contrast at borders electronically. But the concept was first used by painters hundreds of years ago.
One of my favorite portrait painters is John Singer Sargent and several of his larger than life works hang in the National Portrait Gallery in DC where I live. From across the room the detail and sharpness is outstanding. But viewed from 18" you see he quite literally trowelled on the paint with a palette knife rather than a brush. He created the illusion of sharp 3D edges by adding strategically placed white "specular" highlights on objects creating the illusion of sharpness from a distance. We sense 3D shape in 2D art media from contrasting highlight on the top of objects and the darker shadows the directional key lighting, if used, casts.
The reason art forms like line drawings and silk screens work is because beyond reading distance where fine detail can discerned the brain forms it's clues about 3D shape from the contrast on different planes of objects. The brain sees the contrast pattern and matches it to patterns previously seen by eye and the object is recognized from those "primitive" geometric 3D shape clues. That's why in 3D computer art even when a face rendered in wireframe of broad planes of contrasting tones (e.g. Ironman) is recognizable as a human face.
How does that apply to USM choices? How the content is lit creates different tones on the planes and curves of 3D objects. A cross-lit scene will stand up better to enlargement for a huge print than a flat iit one becuase cross lighting creates contrasting planes on objects.
For example if you were to make a print to hang as a mural on a wall a sailboat on the ocean with it's large geometric shapes would be a better choice perceptually and a forest scene because even at low resolution the viewers would recognize the content in the first from the shape of the hull and sail but the forest would look like a green blob.
Reproduction of fur and feathers is also an interesting study in lighting and USM. Both have tiny mirror-like facets which catch light and create sparkle in direct collimated light. It's the sparkle off the individual hard hair shafts that create the illusion of soft 3D texture of the overall body. But the combination of RGB senor processing and AA filter turn those specular shape clues into mush at capture and the illusion of 3D is lost. Using diffuse lighting (overcast day) makes matters worse because diffuse light doesn't create sharp specular reflections. As a result in a photo a dog whose coat sparkles when seen by eye will look dull and matted in the photo due to the lack of specular clues.
Taking a page from Sargent's playbook what I do with animal photos is selectively over sharped the fur to the point where the halos create on borders when over sharpened create the same illusion of 3D texture seen by eye from the real specular highlights. This visitor to my front porch on an overcast day is an example of that technique:
(500, .2, 0) USM:
Second (500, .5. 0) USM layer selectively applied with masking:
I'll also use it for scenics where there is sunlight glinting off water to bring back the sparkle the camera sensor and AA filter kill. So USM decision are best make on a case-by-case basis by comparison based on how the file is output. It only takes a few tests on typical subjects to get a general idea of what works best for you.
Portraits have different USM goals for skin (make it smooth) and eyes and other details (keep them sharp). The solution to that apparent dilemma is solved by using layers in Photoshop. I use a DIY action that copies the base image layer 3x and applies surface blur to one copy, high pass filtering to the second and 500,.2,0 USM to the third. Then using masks to apply and the sliders to adjust I blur the skin, bring back some of the texture with the high pass layer and enhance the sharpness in the eye and mouth area all selectively. Sounds complicated but in practice it only takes a couple minutes per photo.
I'm "old school" and for general USM to overcome the loss of capture resolution I still use the basic USM controls, but with a "twist" suggested in a old Dan Margulis article in a printing trade magazine I read back before digital photography for using USM to sharpen RGB drum scans. Margulis later wrote several good Photoshop books. For screen images I first apply USM = (500,.2,0) in Photoshop, then follow it by Edit > Fade (luminosity). The second Fade step allows before / after comparision with the 0- 100% slider. Changing mode from (normal) to USM produces results similar to sharpeing just the L channel in an Lab file while still in RGB mode.
If you try it you'll likely find most subjects are over-sharpened at (500, .2, 0) but it will vary with how much the screen image has been reduced in pixel size and how much detail the content has. I just move the Fade slider back and forth from 0% (no USM) to 100% (Max USM) and find the point in the middle that looks best by eye.
For prints because the mechnics or printing, size and viewing distance are all variables the best approach for critical work is to take a section of a photo you plan hang on the will, scale it up then cut out a section and save it as a separate file. Then take that section and apply 8 different levels of USM to it and combine those files on a single test print. Put the test print on the wall, step back to different viewing distances and you'll see how USM amount affect perception of sharpness on the borders. As with the Fade slider on screen images it's the side-by-side comparison which will make it easier to determine what works best for that subject / size/ distance on that printer.
Using the Wall analogy as Amt. is increased the "wall" is made higher. Radius, which originally referred to the width of hard printing dots in photo lithography, controls the width of the "wall". I find large amt. 500 and small .2 radius works best for screen images, but with the fade adjustment the amt. actually winds up in the 300 range for most images. For prints I will increase the radius to 1 - 1.5 to account for the difference in printing vs. screen output and then adjust Amt. based on content with a printed comparison test.
Since the mechanical aspects of printing affect resolution you may find different levels of sharpening is needed for photo prints vs. ink jets. You can't judge USM for prints on screen precisely until you have a baseline of experince with your printer and for critical work you plan to print large the time and expense of a comparison test print hung on the wall and seen by someone else more objectively is worth the investment. When in doubt I'll put a test print on the wall and ask the wife which version she likes best and why.