It really depends on the size and the "blur factor" of the original pic.
Usually I use minimum 100%, 1 pixel, 2 threshold level to max 200%, 2 pixel, 2 threshold level. There is something called local contrast usm which is used to improve contrast (kinda), settings at about 400%- 500%, 0.1-0.2 pixel, 0 threshold.
There are no hard and fast rules and recommended settings. You have to try out and convince yourself that the image is sharp by testing on the different settings. Take extra care not to over USM, cos the image may look 'artificial'.
If you are going to develop the photos in a shop, do not perform USM as i was told that the machine are programed to sharpen up the photos before printing. If you have done the USM yourself, tell the shop to bypass/inactivate that step, else you will USM it twice, and hence the artifacts.
Clive, your setting are too aggressive for A4.
Typically I set the amount to 80%-150%, you slide the % from 80% onwards, until you see the grainy/pixel effect on monitor then you stop there.
For raduis typically I set to 0.5-1.0, normally I am just lazy and fix at 1.0.
For threashold I set to 1.
ah..this setting is closer to what i originally guessed that would work for a print of size between A5 and A4 :light: . [my test print verified that 190/1.7/3 is indeed somewhere crossing the "upper limit" already ie too sharp ;-) ]
A *generic* USM setting for web photos (640x480 or less) is:
the 300/0.5/1 for the case of without in-camera sharpening...sounds logical..looks like the main difference is the first variable (ie the %) :think:
"sharpen after resize"...that means the effects of sharpening is non-linear w.r.t. to the change in image size. this sort of ties in with my suspision: while brushing my teeth, i roughly did soem mental maths...i dunno exactly how the sharpening algorithm works i terms of the actual maths involved; but i hazarded a quick guess--suppose the effect of sharpening, denoted by S (higher value of S =>sharper), is dependent on 3 basic veriables: percentage (%), radius (r) and threshold(t). ie we can define S = f(%,r,t). now the main experiment is to find out how S varies with image size I. ie there exists a relationship between S and I such that S=F(I). or in other words, we can say f(%,r,t)=F(I). rearranging, we get:
I = g(%, r, t). where g is a nolinear function. in more detailed terms,
L1.L2 = h. (%)*c1.(r)*c2.[k*(-c3.t)]
where L1,L2, cz, z=1,2,3; %, r, t are all positive with t being the only variable that can be zero (introduction of base k is to remove the singularity of the function g in the case where t=0)
now...the aim is to unravel the values of c1 ,c2 and c3, and most importantly, what the mystery parameter h is. it can either be a positive real number or a function. once we know all these 4 values, then given any image size of L1 by L2, we can input into the function g and derive a solution set of desired values for %, r, and t. that is , assumign that the assumption of the existence of function g is valid i nthe first place, which i believe to be so
I have recently done some research on USM (after I asked Sarah Silver and she didn't know!) - it appears that you should not use a pixel size greater than 0.5 else it introduces too much noise into your picture.
Clive is correct in that you should resize before sharpening so save a copy of your file unsharpened incase you need to resize and then you will want to sharpen again.
The one point I read conflicting views on is whether or not you can apply USM more than once? I have been experimenting with applying first USM at 200/0.5/0 and then a second at 100/0.3/0 to give the sharpness I require. However, I know others who will say that you should not apply USM twice.