Multiple as Opposed to a Single Sharpening

[Steven Thompson]

In Digital Basics Arthur Morris says that he usually sharpens an image in 3 passes using 125/.2/0 each time. He also says that some people prefer to use a single pass at 350/.2/0. I wonder if someone could explain the differences between these 2 approaches and the advanatges/disadvantages of one approach over the other. Thanks. Steve

[Roger Clark]

Steven,
You may want to read this thread:
http://www.birdphotographers.net/forums/showthread.php?t=18534
See, for example, my posts beginning Aug. 30, 2008.

To answer your question above, I don't believe there would be much difference in running 3 passes with the same parameters versus 1 pass with the amount increased 3x. But not knowing the exact algorithm, I could be wrong. Many photoshop algorithms use additive approximations for multiplicative processes, so 3 applications could result in greater error. It also would have greater round-off error with multiple applications with the same settings, especially on 8-bit files.

Note, as the referenced thread discusses, unsharp mask does not actually sharpen; it changes accutance.

There is a scientific reason for applying sharpening algorithms in multiple passes, but in each pass with different settings, and that again is discussed in the previous thread.

[Roger Clark]

OK, after thinking about this one overnight, I have to qualify my response above regarding multiple versus single pass sharpening.

If the algorithm does additive math, then there would be no difference in single versus multiple passes. For example if a part of the image were increased in intensity 10% per pass, an additive algorithm for 3 passes would give
1.1 + 1.1 + 1.1 = 1.3.
If it is multiplicative then it would be:
1.1 * 1.1 * 1.1 = 1.331

In a multiplicative algorithm, 3 passes would be like cubing the effect not simple 3 times more. A more extreme example:
2 + 2 + 2 = 6 versus 3*2 = 6 versus 2 * 2 * 2 = 8
4 + 4 + 4 =12 versus 3*4 =12 versus 4 * 4 * 4 = 64
and as the value goes up the difference in in multiple passes increases. It would take some controlled testing to see what unsharp mask does in photoshop.

One problem with multiple passes that I can envision is that with a multiplicative algorithm, the effect becomes greater the larger the difference. So in unsharp mask, multiple passes could cause more halo artifacts on higher contrast edges when little change is seen on low contrast edges if the algorithm is multiplicative. (Again, an additive algorithm is often used as an approximation multiplications to improve speed.)

[Jonathan Michael Ashton]

Boy you guys have started something now! Could anyone advise me over the optimal way of sharpening for the web - I know this has been covered in different threads before but I am coming from a slightly different angle.

I always shoot RAW and recently I use DPP so I end up with a TIFF in Photoshop CS2. Now if I am going to make an image for a print I make a large one and set to 180dpi as I use Epson printers. Now assuming I am not going to make a print but I am going to produce an image for BPN would it be best practice in the first place to crop or Image size to for e.g. 800x 1042 and 72dpi, do all the Photoshop buts and then sharpen - (I usually use Smart Sharpen) as opposed to leaving a full sized image, then Sharpening, reducing size, sharpening, reducing size to 800x1042 and then Smart Sharpen? Any comments would be gratefully received.

Jon

[Alfred Forns]

Thanks Roger that makes sense !!!

I remember Tim Gray recommending against multiple passes but there was no explanation.

[Alfred Forns]

Hi Jonathan I getting a little confused but this is what I do for posting

Do all the PS work at full size then convert to sRGB, reduce to posting size using bicubic sharper then apply your sharpening.

I use the capture sharpening pre set in LR, its very similar to PhotoKit Sharpener (import) will give the image look slightly better while evaluating.

Since you are in the subject maybe Roger can clarify a point .... I thought using Genuine Fractals would help when having to upsize an image but in one of the questions and answers Tim gets he said it was just as well using bicubic smoother?

[Jonathan Michael Ashton]

Thanks Al

Jon

[Steven Thompson]

Thank you all. From this string I would conclude that there is no advantage to multiple unsharp mask passes using the same settings and, in fact, there may be some disadvantages. Let me pose a second question - What clues do you look for to recognize an oversharpened image? I'm talking about the fine line between a sharp/crisp photo and one that is just a bit overboard. If I tinker with the settings for a single unsharp mask I would just would like to have some idea when I have gone too far. Steve

[Fabs Forns]

An over-sharpened image has shine and a frozen look to the details in feathers, not to go into halos and artifacts. I suggest moving the slider to the right till it looks horrible, then moving it left until it looks normal.
Over-sharpened images had a "frozen in time" look to them.

[Roger Clark]

Hi,

The methods I use and the order are documented at:
http://www.clarkvision.com/photoinfo/digitalworkflow

Whenever I downsize, I usually do a little unsharp mask. Generally for an image for the web, which is a large downsize, I'll do something like radius = 0.2 to 0.4 and amount around 60.

I found that the amount of unsharp mask that produces visible artifacts depends on the monitor. For CRT monitors, you can do more unsharp mask and not see a bad effect but that same image on and LCD may look over sharpened with halos. As a growing number of people are viewing images with LCDs, if you are still using a CRT, you might want to get an LCD to see how others might be viewing your images. Of course most of those LCDs are uncalibrated with too high a contrast and saturation--yuk.

[Robert Amoruso]

Thanks for the informative thread and read. Makes me think to revisit my methods.

[Jonathan Michael Ashton]

Noted Ron, thank you, I keep on picking up little gems here and there, you have reinforced some of my conceptions.

Jon

[Steven Thompson]

Thank you all again, this is very helpful. Steve