Why only unsharp masking for sharpening?

[revi]

Hi,

I've been lurking around here for a while now, and when sharpening is discussed, it always seems to be about using unsharp masks. Why is this the only technique used?

I use Digikam under linux for most editing I need (at least for the global image corrections, it is one of the few programs I found under linux that allow raw 16-bit editing). There I've got the choice of three sharpening methods, one of which is a deconvolution method, that is slower but gives better results (I think): it preserves more of the fine detail than the other methods (here is a description of the three sharpening methods).

Note that I'm only asking about sharpening here, not about improving local contrast through USM.

Remco

[Colin Southern]

when sharpening is discussed, it always seems to be about using unsharp masks. Why is this the only technique used?

Hi Remco,

For me the answer is simply because it's effective, but quick.

[rick55]

Hi, Remco;

Deconvolution is for a different problem than people are usually applying unsharp mask (USM) to. USM is the tool of choice, as far as I've seen, for the sharpening in the "normal" workflow, which is discussed in this tutorial.

Deconvolution is more for rescuing an image that's badly blurred, which isn't something I've seen discussed here a lot. The capability exists in Photoshop, under smart sharpen, and in external tools. This thread discusses some of the issues with it. It explains why it probably isn't a good choice for an image that's pretty good to start with.

Cheers,
Rick

[revi]

Thank you, Rick

I missed that link, and it is very clear What got me wondering originally is that USM doesn't really sharpen the image, whereas deconvolution does/can. It is true that deconvolution tends to increase noise, but careful use gives a nice result; a bit softer than USM, but I tend to like that

[flipmode]

Hey Remco,

You can also use high pass filter with overlay or soft light blend mode for sharpening. It's highly referred to these days in many articles and online. Did you know you can also sharpen using surface blur. Check out this tutorial video about midway down the page.

http://www.scottkelby.com/blog/2010/archives/8691

[Colin Southern]

What got me wondering originally is that USM doesn't really sharpen the image)

Huh?

[rick55]

Hi, Remco;

As described in the Sharpness Tutorial, sharpness is the combination of acutance and resolution. Acutance is how fast image intensity changes across space. If an image has very quick transitions, going light to dark or dark to light very few pixels apart, it has high acutance. This is what USM affects. High pass filter does, as well, as Chris says. High pass refers to frequency domain representation of the image, and higher rate of change of intensity per unit distance is higher frequency, and will "pass through" the high pass filter.

If you want to limit the definition of sharpness to resolution, and say that a deconvolution operation can restore (within a lot of limits) some resolution that was lost with focus blur, then in that sense, yes, deconvolution provides an aspect of sharpening that USM doesn't.

But resolution and acutance are two separate measures of an imaging system.

Cheers,
Rick

[revi]

Hi Rick;

That's what I was aiming at, except that I didn't make the distinction between acutance and resolution explicitely.

I'm by no means an expert, but from what I see on my images, even a moderate amount of deconvolution sharpening gives me some extra fine detail (=resolution), where USM actually coarsens the details a bit. The masking method can then make this stand out a bit more (=acutance).

I'm not trying to get anyone to switch to other methods, just trying to understand why they are used so little (that seems due to a combination of USM is fast and it works). Especially as there's whole books written about sharpening, and there are some methods that also require quite a bit of work/fine tuning, it seems strange to use only one or two basic methods.

For images that are worth it, would the following be a good workflow?
- capture sharpening with deconvolution (resolution enhancement)
- creative sharpening (increasing acutance)
- output sharpening (increasing acutance)

Given how the different methods work, I wouldn't use deconvolution after USM...

[Colin Southern]

it seems strange to use only one or two basic methods.

In my case I consider anything more time-consuming to fall under the "law of diminishing returns".

[Steve S]

Hi remco, i've been having great results using smart sharpening in photoshop. I combine it with an edge mask though.

[rick55]

Hi Rick;

That's what I was aiming at, except that I didn't make the distinction between acutance and resolution explicitely.

I'm by no means an expert, but from what I see on my images, even a moderate amount of deconvolution sharpening gives me some extra fine detail (=resolution), where USM actually coarsens the details a bit. The masking method can then make this stand out a bit more (=acutance).

I'm not trying to get anyone to switch to other methods, just trying to understand why they are used so little (that seems due to a combination of USM is fast and it works). Especially as there's whole books written about sharpening, and there are some methods that also require quite a bit of work/fine tuning, it seems strange to use only one or two basic methods.

For images that are worth it, would the following be a good workflow?
- capture sharpening with deconvolution (resolution enhancement)
- creative sharpening (increasing acutance)
- output sharpening (increasing acutance)

Given how the different methods work, I wouldn't use deconvolution after USM...

I don't see anything wrong with this workflow, based on anything I've read. Most of my reading on sharpening has been here, and "Real World Image Sharpening," by Fraser and Schewe. I've also done 1D (audio) digital signal processing at work, and I've done a little bit of 2D signal processing for my own purposes. But my experience in working on images using off-the-shelf tools is limited.

But I don't see any advantage, unless the lens/focus is limiting the resolution. If the lens is good, the limit is going to be the anti-aliasing filter, I think. This is a low-pass filter, so I think a high-pass filter masked at the high-frequency edges is probably a pretty good choice to reverse the effect. The effect will tend to flatten the overall transfer function of the system.

Cheers,
Rick

[Pictus]

G'MIC is a GIMP plug-in with Richardson–Lucy deconvolution
http://gmic.sourceforge.net/index.shtml

For Windows, Topaz Detail have an interesting sharpen algorithm and also deconvolution, Akvis Enchancer deconvolution is also very good, Focus Magic and I guess Nik Sharpener does too, but I an not sure about Focal Blade.

To use Photoshop plug-ins in GIMP http://tml.pp.fi/gimp/pspi.html

Interesting http://members.ozemail.com.au/~binar...ersharpen.html

Some tests http://www.hugorodriguez.com/index_nitidez-perfecta.php

[danattherock]

Interesting thread and I agree with the OP. I have heard about unsharp mask so many times, makes one think there are no other options. For anyone so inclined to answer, would I be best served sharpening in Lightroom 2 or using my new Nik plug-in (Sharpener Pro 3.0)? I have Lightroom 2 and the Nik complete collection of 5 plug-ins, but this is the only post processing software I own. Not sure which I should be using for sharpening. Got LR2 a few weeks back and the Nik kit a few days back. Any suggestions appreciated.

http://www.niksoftware.com/sharpenerpro/usa/entry.php


-Dan