Unsharp mask is redundant!

[Stuart Hill]

Just thought I'd share a little trick I've picked up from a retouching forum. Apparently USM is inaccurate in the way it sharpens (have to admit, this is from high end beauty techs!). The trick is to 'split' the image into detail layers and colour layers. It's quick once you've done it a few times.

It opens up real possibilities when say, retouching a background, much easier. Worth a play, please try it and let me know your thoughts.

here goes:-

open your image in PS, 16 bit (must be 16bit or this way won't work)
make 2 copies of your original layer.
working on the second layer (background copy) run a gaussian blur. (the detail you blur here will become your sharp layer. Try different strengths. I start at 4.0 or, 1.0 for the blur see what works for your image)
working now on the top layer (background copy 2), go to IMAGE (main menu bar) and down to APPLY IMAGE.
In this dialogue box change the 'layer' to the blur layer (background copy), check the INVERT box, change the blending mode to ADD, SCALE to 2 and OFFSET to 0, then hit OKAY.
Still working on your top layer, change the blending mode to LINEAR LIGHT
And thats it, you now have your colour data on one layer and the detail or texture data on the other. But nothing has changed you say!

Well, try duplicating the top layer and see what happens.

What this means is, you can dodge or burn, change colors, curves, contrast, all with out worrying about losing detail because the detail is on a different layer.

Hope this all makes sense, if not, mail me and I'll mail you back an action.

regards.
Stu.

[John Chardine]

Thanks for this Stuart. I was aware that you could duplicate what USM does but I'd never actually done it before. Still not sure exactly how it works. It's not clear to me how this manual USM method is any more accurate that the Ps method. I know USM is an old method that was used in darkrooms.

The comment about separating colour and detail intrigued me because this is exactly what the LAB colour space does. Chroma (colour noise) is in the colour channels- A and B- and Luminance noise is in the Luminance (L) or detail channel. Knowing this allows you to reduce noise and sharpen images much more effectively than in colour spaces like RGB, where detail and colour information are intermingled.

[Stuart Hill]

Thanks for trying it John. Just experimenting with it myself to see just what can be done with the technique.

would advise anyone interested to have a read:- http://www.modelmayhem.com/po.php?thread_id=439098

Quite cool. By making different "detail" layers and using layer masks you can smooth or sharpen at different strengths non-destructively. Don't think its a batch kind of action but for a special image?

regards.
Stu.

[James Shadle]

Thanks Stuart,
I'm always open to new ideas.

[Robert Amoruso]

Thanks for this - cool trick. I think I read something like this in a book recently but can't recall exactly where it was.

[Roy Churchill]

Thanks for that Stuart. I have tried it and it works, I don't quite understand how and why. What I really like about it is the perceived sharpening without a hint of a halo But I do not seem to be able to control the amount of 'sharpening'. The examples I tried seemed overdone. I have tried altering the amount of Gaussian blur but that does not seem to make and difference - am I missing something?

[Stuart Hill]

Glad you found it a useful tool. You're not missing anything Roy. It is quite a complex theory all based on maths within photoshop. Really is worth reading the link near the top of the thread, though be warned. It hurts your head after a while.

I've found if you blur the sharp layer it effectively lessens the sharpening. As I said. This is a technique I'm still studying but will post any future gems here.

Please continue to share your progress and findings.

Kind regards.
Stu.

p.s. Try using other blend modes for the sharp layer. Soft light, vivid light to give more or less vigour to the sharp layer. Also you can drop the opacity of the sharp layer down from 100% to 50%.

You can also drop the FILL down to 50% (under the opacity setting on the layer tab) not sure what the difference is though.

[Gordon Lindsay]

Thanks Stuart. I have tried it and it really works, I copied the top layer to make the effect "oversharp" then added a mask to mask out or reduce the amount of "oversharpening", then made an action from it and its a very easy way to enhance details. Thanks again for the idea.

[Roger Clark]

Hi,
This thread came up while I was on travel. The method is an interesting exercise in unsharp masking. The method basically follows the math used in unsharp mask, but because of the many steps it seems to me you have less control. I find it easier to change to LAB mode and unsharp mask on the luminance channel. Unsharp mask originated in film days. One would make a negative of your negative (a positive) that was slightly out of focus and underexposed. Then sandwich it with the original negative and print it (or make another image on film (a positive) then make a new negative. So in the method described in this thread, the blur step with the check the invert box is like the underexposed blurry positive. The add is like the sandwich of the two. These same steps are in the unsharp mask tool. And, as we have discussed before, unsharp mask does not actually sharpen; it modifies edge contrast (accutance) which gives the illusion of sharpening. Talk about less control: imagine exposing, developing and drying your mask, then aligning and exposing more film? Took many hours per iteration!

There are more variations on the unshapr mask theme, including find edges, and then apply the unsharp mask to the edges. One could do that on the luminance channel in LAB mode too.

For those who have not seen it, check out the sharpening information in the educational resources forum:
http://www.birdphotographers.net/for...ng-Information!

A side note, I did speak to the software developer for one of the astronomical image processing programs that includes Richardson-Lucy deconvolution and he thought he could make a low cost version of the program that only did Richardson-Lucy deconvolution for photographers.

Back to unsharp mask, I would encourage everyone to try sharpening the luminance channel in LAB mode. LAB is the widest color space, and when you sharpen the luminance channel, you do not get color shifts on edges. Also, if you use the curves tool much, try it on the luminance channel, again, no color shifts.

Roger

[Stuart Hill]

Hi Roger, was hoping you'd see this and wanted your input. If you get time, would it be possible for you to have a quick read: http://www.modelmayhem.com/po.php?thread_id=439098

Some of the guys are pro retouchers and state ps usm is not accurate (though they mention the Richardson-Lucy decon is best). Would like to hear what you think of the discussion on the above thread.

For using curves tool, would changing the blending mode to luminosity not have the same effect in RGB as working a curve in LAB luminance channel?

Appreciate your input here Roger.

Kind reagrds.
Stu.

[John Chardine]

When I saw this thread I was hoping Roger would see it!

I routinely now sharpen the Luminance channel in LAB colour space. In general this space has not been embraced by photographers so I'm not surprised to see alternative processing paths advocated by high-end retouchers.

[Roger Clark]

Hi Roger, was hoping you'd see this and wanted your input. If you get time, would it be possible for you to have a quick read: http://www.modelmayhem.com/po.php?thread_id=439098


Some of the guys are pro retouchers and state ps usm is not accurate (though they mention the Richardson-Lucy decon is best). Would like to hear what you think of the discussion on the above thread.

Hi Stuart,
OK, I spent a lot of time this evening going through the above thread. Probably the most interesting thing to me was the fact that most avatars are glamor shots of women, but many (most?) of the posters are guys.

Anyway, let me start by saying that photoshop was born on slow PCs many years ago. Machine are thousands of times faster. Photoshop in the beginning, up to recently and maybe even still true with some toolls, did some approximating tricks to make the tools faster. For example, is an algorithm called for multiplication, photoshop did an add. This works if the application is small and the error is not large. The contrast tool was a good example of this. And even today, some tools, like increasing saturation with the hue/saturation toll produce simply bizarre effects when pushed very far (CS4). So it wouldn't surprise me if unsharp mask were shown to be inaccurate.
I could go on about photoshop's funny 15-bit math on 16-bit files, but we need not go there.

But in the above thread, I never saw it established that unsharp mask actually produced inaccurate results.

Adobe has been improving the math, so some statements about inaccuracies need to be tied to specific versions of PS.

Then is unsharp mask is too inaccurate for your application, why not use smart sharpen, which does have improved accuracy, and different blur mmodels, including Gaussian?

Now I'm not saying the method cited doesn't work. It does, just I wonder about the effort and control one has compared with other photoshop tools. There are many ways to do things in PS (I think that was said in the above thread), but why bother if it takes longer to get the job done?

For using curves tool, would changing the blending mode to luminosity not have the same effect in RGB as working a curve in LAB luminance channel?

If you mean duplicate a layer and change the layer mask blending mode to luminosity and then running curves on that layer, yes it appears to do the same thing.

Note, however, LAB mode is a larger color space. Actually, it is the largest color space of all. As John says, photographers seem to have not embraced this color space, which I find strange. They debate Adobe RGB versus Pro Photo RGB, both of which are small compared to LAB.
http://en.wikipedia.org/wiki/Lab_color_space

With such a large color space, 16-bit data (or larger) should be used.

Roger

[Norm Dulak]

How does "Smart Sharpen" differ from USM? Does it overcome the real or perceived difficulties with USM? :confused:

[Don Lacy]

Note, however, LAB mode is a larger color space. Actually, it is the largest color space of all. As John says, photographers seem to have not embraced this color space, which I find strange. They debate Adobe RGB versus Pro Photo RGB, both of which are small compared to LAB.

Roger, Some of the reasons are given here http://www.luminous-landscape.com/fo...?topic=41816.0 would be interested in your opinion on their argument that the benefits of working in LAB is offset by the lost of data that occurs from switching back out of LAB for output.

[Stuart Hill]

Thanks Roger for taking the time out and posting back your thoughts. It really is appreciated. I've always been slightly scared of LAB. Even though I've been using ps since the beginning I've never ventured there. Splitting the info has made retouching images a lot easier.

After hearing your thoughts, will investigate the world of LAB.

kind regards.
Stu.

[Roger Clark]

Roger, Some of the reasons are given here http://www.luminous-landscape.com/fo...?topic=41816.0 would be interested in your opinion on their argument that the benefits of working in LAB is offset by the lost of data that occurs from switching back out of LAB for output.

Hi Don,
My first reaction to the above thread is that its typical geeks (hey I'm one too) over-discussing things that don't matter. A few simple experiments will show how much is lost on conversion to LAB and back.

But before I cite some numbers, take a 16-bit file and read it into photoshop and write it back out as a different file name. If you have a real image analysis system, you will find that photoshop changed the values! So simply reading and writing a 16-bit tif file in photoshop is destructive by one to 2 bits.

I took an 8-bit file, converted it to LAB and back ro RGB, then copied it in a new layer over the original RGB image, changed the layer mode to difference, flattened the image and analyzed the result. Values changed by one DN.
So yes, there is a slight loss, but not the 22 that is stated in the thread above.

I did the same thing with a 16-bit file. I found changes about 1 part in 10,000. Now, the signal-to-noise ratio of any high end digital camera is maximum of about 300, far less than 1 in 10000. All current digital cameras collect less than 100,000 photons per pixel (80,000 is the most I have seen). For a loss in 1 part in 10,000 one would need to collect 10,000 * 10,000 = 100,000,000 photons for the noise to be small enouigh for that loss to be significant, or 1000 times more than any camera can currently collect. Bottom line, any loss is negligible and not much different than reading and writing a tif file in photoshop.

A side note: the idea in the thread above that parametric raw conversion is totally lossless is fantasy. Raw conversion will lose by integer truncation, and artifcats will be introduced because the interpolation algorithms are imperfect. In practice, the signal-to-noise limits of real camera data make the argument silly.

Thus, any loss in conversion of a 16-bit image is very negligible, and probably not a big factor in many 8-bit images.

I'm not a fanatic on color accuracy. The reason was alluded to in some of the discussion on the LL thread. First, color is defined relative to the 1931 study. Fine, but that is only one model. Another model gives different results. Why? under different conditions, our perception of colors changes. One could in theory only have accurate colors for all perceived colors with a very specific light source (spectral response and intensity), specific inks and paper, and the print illuminated at a given angle and distance, and viewed at a given angle and distance. Change any one of these and colors will shift, and much more than in any one model (like LAB or pro photo RGB) would give. The obvious demonstration of this is view a print in an office in fluorescent light, at home in incandescent light, versus outside in sunlight, versus outside in overcast light. In each case perceived colors will change. Which is most accurate? The real world is quite complex. One can do reasonably well with a color managed workflow and get beautiful prints, or beautiful displays on a computer monitor. But to make everything highly accurate is a virtually impossible goal.

I think its better to spend one's time taking pictures than debating and over-interpreting details of color spaces.
---Just my opinion.

Roger

[Don Lacy]

My first reaction to the above thread is that its typical geeks (hey I'm one too) over-discussing things that don't matter

Hi Roger, Thanks for taking the time to reply I thought the same about half way thru the thread it seemed to me that the guys who literally wrote the books on color management and ACR had a bigger beef with the guy who wrote the book on using LAB in PS then they did with the actual process.

I think its better to spend one's time taking pictures than debating and over-interpreting details of color spaces

I think that you are right yet again:D

[Tom Graham]

repeating Norm D. question -
"How does "Smart Sharpen" differ from USM? Does it overcome the real or perceived difficulties with USM?"
Was answered? I didn't see it, also like to know.
Tom

[Roger Clark]

Tom,

See the thread:
http://www.birdphotographers.net/for....-Unsharp-Mask

Roger