Digital images are inherently unsharp due to the way they are captured on the sensor on discrete red, green or blue sensor sites then converted into combined RGB pixels. There is a "anti-aliasing" or "AA" filter over the lens which blurs the to prevent a stair-step pattern on diagonal lines in the scene.
Sharpening can occur at several places in the workflow. If shooting JPGs it is applied in the camera and adjusted via the camera menu. If shooting RAW the sharpening can be applied, or not, at the RAW editing stage.
There is no universally correct way to sharpen an image. In the end it comes down to a personal subjective judgement for each individual image at the size it is reproduced and the distance at which it is viewed. The perception of sharpness and contrast between objects is inter-related. In an image viewed from 18" the eye can resolve fine detail and can detect a lack of sharpness in fine detail. But as the viewing distance increase the eye loses its ability to resolve fine detail and instead reacts to contrast differences to create the illusion of shape. That is way reproduction size are viewing distance are important variables in the USM equation. As an image gets larger and viewed from further away it needs different sharpening than one intended for eye level viewing. The same is true of very small images, such as those used for thumbnail icons, which often look better if oversharpened.
The origins of the name and technique
It may seem odd that a technique which sharpens and image is called "Unsharp" masking. The process has it origins in the days when CYMK color separations for printing were made by screening a color print or transparency into separate Cyan, Yellow, Magenta, and blacK dot patterns with halftone screen and a slightly out of focus or unsharp positive mask was added. At some point someone discovered, likely by accident, that when an slightly unsharp positive image of the original was added to the sandwich at this step the apparent sharpness of the reproduction increased.
How it Works
Unsharp masking increases the contrast between different tones and colors in an image, but only in a very narrow band on either side of the border. The darker side is made a bit lighter and the light side a bit darker.
By way of analogy is would be like building a brick wall on one side of the border between two countries and digging a deep trench on the other side. As with USM in an image, we'd notice more when that border was crossed.
The image isn't really any sharper in terms of actual resolution but the eye is tricked into thinking it is, just as we are tricked into accepting that a pattern of light and dark in a photo is a 3D object. The contrast added at the transitions creates the illusion of more sharpness, but overdone the lightening effect on the darker side will create visible white halos around the darker objects, the sign that the image is over-sharpened.
The Evolution of USM
The term "unsharp masking" was carried forward into the era of analog and digital drum scanners, where it was produced electronically by programming the scanner to detect tonal transitions and then add a bit more or less halftone dot size that would normally be needed. The term "radius" was used to describe how many dots on each side of the transition the sharpening was applied to during the scanning process. A radius of 1 equalled half the width of a halftone dot. The term "threshold" was used to describe how much contrast between tones was necessary before sharpening action was applied.
Photoshop adopted and carried forward the names "unsharp masking", "radius" and "threshold" because the first users of Photoshop were color separators familiar with those terms. A third value "amount" was added to control how aggressively the USM was applied, getting us to where we are today where USM is defined by three numerical values in the format: Amount, Radius, Threshold.
The analogy of wall and trench is useful in understanding what role the three controls play in the process:
AMOUNT: The first number represents the amount of sharpening.... 0 = none 500 = maximum Think of this as controlling how high and deep the walls and trenches at the tonal borders are.
RADIUS: The second number represents radius... In the days of drum scanner separations "radius" actually referred to the radius of one halftone dot. In Photoshop it refers to one pixel. Think of this as controlling how wide the walls and trenches at the tonal borders are.
THRESHOLD: The third control is threshold. It controls when the sharpening is applied. 0 = everything is sharpened. The greater the threshold the more contrast or difference there must be before sharpening is applied to that area. For example a setting of 5 would sharpen the dark contrasting eyes and eye lashes but not apply sharpening to even toned skin.
The Ying and Yang of Sharpening
Amount and Radius are the Ying and Yang of sharpening. When one is large the other must become smaller to maintain balance and prevent over-sharpening.
If a small radius such as .2 is selected with very large amount, it is like building a tall but narrow wall. Large amounts of sharpening can be applied because only a very narrow area on either side of the tonal borders is affected, too small to create any distracting halos at normal reading distance. For example when 500, .2, 0 USM is applied the image will be much sharper looking. I use those settings 500, .2, 0 for my screen images. But if the radius is increased only slightly to 500, .3, 0 the image may appear over-sharpened. But if amount is reduced until the halos disappear and the image will again appear sharp but not over-sharpened.
500, .2, 0 or 350, .3, 0 of 150, 1, 0 may all wind up looking exactly the same to your eye exactly the same on your monitor. At the microscopic pixel level the effect is created in two different ways -- tall and narrow vs low and wide -- but the net effect is the same. That in part is why there is so much disagreement regarding which method is best.
Thresholdis the third USM control in Photoshop controls where in the image USM is applied. The determining factor is the amount of difference between two different colors or tones. If threshold is set to 0 everything in the image will be sharpened. As the value is increased only areas with higher contrast with the surrounding area will be sharpened. Beyond a setting of 5 every little in the image will be sharpened. I find it is best to start at 0 then observe how the image changes when slowly increased to about 5.I let my eye tell me what threshold looks best.
Two schools of thought on USM emerge
Because of the Ying/Yang of USM its only possible to use either a large amount with a small radius, or a moderate amount with a large radius. Most users choose one or the other, some becoming rather zealous in their belief that theirs is the one true faith...
Having come from a background of making halftones and color separations on process cameras and scanners and the having to print them on a web press long before first using Photoshop (1987) or actually owning my own digital camera (2000) I have come to understand both how an image is changed when printed and how the human eye will perceive those changes.
When a RGB file is printed it is first converted into its CYMK equivalents and then converted into a gazillion tiny ink droplets which get spit onto the paper overlapping each other randomly. When the ink hits the paper it will soak in spread to some degree. In the process an image which might look razor sharp on a computer monitor will, relatively speaking, wind up looking like a fur ball the cat coughed up.
One of the things USM can do for a printed image is to compensate for all that random spitting of ink and how the ink soaks and spreads into the paper. Thus, it would seem logical to sharpen more for printing than for screen viewing. But using which method?
The small radius / large amount method, while excellent for screen images, is a poor choice for printing because it would just pile more ink on the paper to soak in and run together. The wall melts away and fills the trench leaving level ground; and not much apparent sharpening.
The large radius / moderate amount approach works better for printing because the wider "trenches" or white areas where tone has been artificially been reduced to add contrast are less likely to get obliterated by the physical rigors of printing.
Thus I personally have a foot in both camps, using small radius/large amount USM for my screen images and large radius / moderate amount USM when I print.
There Is No One Size Fits All Solution
Sharpening by adding contrast transitions is just an optical illusion which tricks our visual system. The limiting factor for our vision is the resolving power of the eye. As the viewing distance to a print becomes greater, the detail in the print becomes more difficult to see. Using greater amounts of USM for larger prints will increase their apparent sharpness. They may look horribly over-sharpened at reading distance, but when viewed from across the room it will look "normal" or similar to an smaller normally sharpened print at reading distance. As as a rule-of-thumb the bigger the print is the more USM can be applied.
Ironically the same is true for very small copies of images, such as thumbnails used on a web page for linking to a larger copy of the photo. The sharpening which makes the larger print look best will leave the smaller thumbnail looking soft and mushy. Slightly over-sharpening the thumbnail will boost the contrast by adding sparkle to the highlights and make it more visually attractive and effective.
As with lighting there is no "one size fits all" solution for USM; only cause and effect. The way to evaluate USM is not by "pixel-peeping" but rather by observation and side-by-side comparison of the two methods with various amounts and radius applied, with the comparison done at size, output resolution and viewing distance which will be used for the final product. In other words a print for a scrapbook viewed from 18 inches would need to be sharpened differently than the same photo printed 30 x 40 inches for display on a wall from across the room.
Testing Tells the Tale
To determine what So you might print the same photo ten with ten different USM setting and compare which looks best on your printer:
100, .5, 0
200, .5, 0
300, .5, 0
300, .5, 0
500, .5, 0
100, 1, 0
200, 1, 0
300, 1, 0
300, 1, 0
500, 1, 0
It's not rocket science, just cause and effect of two main variables; amount and radius, and they are proportional. If one goes up, the other must come down or else the halos appear.
When halos appear
your image is going to hell
not on its way to looking heavenly.