To sharpen or not to sharpen

[apaflo]

There are a ton of things called 'Sharpen' out there, not all of them use a Gaussian Kernel anywhere.

There's a lot of ways to amplify higher frequencies, or attenuate lower ones, and not all those ways have a thing called "sigma" anywhere in them.

I have repeatedly specified "high pass" and "convolution" methods for sharpen. And have pointed out that most "Sharpen" tools in image editors use convolution. As noted, all algorithms using convolution have a sigma parameter, even if the user cannot set it.

Can you cite a Sharpen tool commonly seen in a typical editor that does not use convolution?

This is complete nonsense. This so-called "definition" for sigma is not only wrong[...]

It's pretty standard stuff. See and also try Gaussian function - Wikipedia, the free encyclopedia and Standard deviation - Wikipedia, the free encyclopedia for definitions and examples.

 

[apaflo]

I'll use that as an example of curves sharpening only.

I'm sorry, but all you have done is change the gamma curve, increasing contrast in some areas and decreasing it in others. That is not what "sharpen" means.

Here is a really good article by Roger Cicala of LensRental.com about sharpening. He does a very good job of describing sharpness, acutance and resolution.

 

[amolitor]

You can and often do convolve with things that are not Gaussian functions. Gaussian things and convolution things are in fact completely separate ideas, that have only occasional and largely accidental overlaps.

Wavelet sharpening.

 

[Derrel]

I'm placing a bet on Gavjenks for $20. I have full confidence that he will win the debate.

Yeah, but the thing is, you just NEVER really "know" for sure...

 

[Rick58]

Just think, once upon a time you only had to worry about grain and a good enlarger lens.

 

[snowbear]

I like turtles.

 

[Gavjenks]

Can you cite a Sharpen tool commonly seen in a typical editor that does not use convolution?

Why yes I can! The curves tool in photoshop... =D
As you yourself pointed out, it is one for one pixel for pixel, without consideration of neighbors. I.e. no kernel/convolution.

Also, unsharp mask done in analog is an example of a sharpening tool that does not use a convolution, in a darkroom. (assuming you mean actually calculated on a computer. It is a convolution in physical space, but if you count that, then everything ever in the universe is a convolution pretty much)

Just think, once upon a time you only had to worry about grain and a good enlarger lens.

Both unsharp masking and contrast masking can be done analog. Could be having the exact same debate in 1975, interestingly.

I'm sorry, but all you have done is change the gamma curve, increasing contrast in some areas and decreasing it in others. That is not what "sharpen" means.
Here is a really good article by Roger Cicala of LensRental.com about sharpening. He does a very good job of describing sharpness, acutance and resolution.

1) (You forgot the link.)
2) That's not a gamma adjustment. Gamma is a specific power equation which is not what I did, and which, importantly, does not sharpen things unlike what I did. Gamma only makes the whole image darker or lighter in a non-linear fashion. Well... I suppose it could SORT OF "sharpen" something if it brings the lightness values from outside the eye's range to inside of it, such that you can see edges you couldn't before. But I'm not sure I would count that.
3) I did actually perform every single step involved in a classic sharpening algorithm. I first applied a sub-algorithm to choose where the edges were. In this case, that sub-algorithm was me manually looking at the histogram and picking what I thought was the pointiest bit. This serves the same purpose as a blurring mask. In fact, I would go so far as to actually say that what I did WAS a (1-dimensional) mask. it's just cruder. I then proceeded to apply a curve in the reverse of the gradient of the edge in order to improve its acutance, which is precisely what the USM tool does. Instead of using a gaussian, however, I simply eyeballed my own curve. But it serves the exact same purpose, again just more organically and not a strict equation behind it.

And you can't say "Oh it's unfair to offload part of the algorithm to your brain, that doesn't count," because YOU DO TOO when you use the unsharp mask tool. Your brain is choosing each parameter using algorithms of its own, before the computer runs the USM algorithm. This is no different, just the brain is doing more of it.

 

[apaflo]

You can and often do convolve with things that are not Gaussian functions. Gaussian things and convolution things are in fact completely separate ideas, that have only occasional and largely accidental overlaps.

Wavelet sharpening.

You are the only one who has suggested a Gaussian function is necessary for sharpening, and I pointed out that is not the case. The reference I gave was to show you how convolution necessarily has a sigma parameter.

Yes, wavelet sharping is an example of a non-convolution form of sharpening! It's one that is virtually always named differently too... Another, which probably will not be obviously identified, would be one using a Fourier transform.

 

[amolitor]

You are the only one who has suggested a Gaussian function is necessary for sharpening

I did no such thing. I defy you to point out where I said such a ridiculous thing.

The reference I gave was to show you how convolution necessarily has a sigma parameter.

Again, convolution does not necessarily have a sigma parameter. You are simply wrong here.

There is not room in the margin of this book to give you a course in digital signal processing, but it is painfully clear that you're operating at least a couple steps past the limits of your knowledge here. It's embarassing.

 

[apaflo]

I'm sorry, but all you have done is change the gamma curve, increasing contrast in some areas and decreasing it in others. That is not what "sharpen" means.
Here is a really good article by Roger Cicala of LensRental.com about sharpening. He does a very good job of describing sharpness, acutance and resolution.

1) (You forgot the link.)

Sorry about that! Here it is: LensRentals.com - Have You Seen My Acutance?

2) That's not a gamma adjustment. Gamma is a specific power equation

The specific power equation is "gamma correction". Gamma is a measure of contrast. Look it up, and almost any dictionary will say it is "a measure of the contrast reproduced in a photographic or television image"

which is not what I did, and which, importantly, does not sharpen things unlike what I did. Gamma only makes the whole image darker or lighter in a non-linear fashion. Well... I suppose it could SORT OF "sharpen" something if it brings the lightness values from outside the eye's range to inside of it, such that you can see edges you couldn't before. But I'm not sure I would count that.

What you did, using a Curves Tool, was change the gamma curve. That's what "Curves" does. And you are correct that changing the gamma curve does not sharpen an image.

"The key concept with curves is that you can never add contrast in one tonal region without also decreasing it in another. In other words, the curves tool only redistributes contrast."
Using the Photoshop Curves Tool

3) I did actually perform every single step involved in a classic sharpening algorithm. I first applied a sub-algorithm to choose where the edges were. In this case, that sub-algorithm was me manually looking at the histogram and picking what I thought was the pointiest bit. This serves the same purpose as a blurring mask. In fact, I would go so far as to actually say that what I did WAS a (1-dimensional) mask. it's just cruder. I then proceeded to apply a curve in the reverse of the gradient of the edge in order to improve its acutance, which is precisely what the USM tool does. Instead of using a gaussian, however, I simply eyeballed my own curve. But it serves the exact same purpose, again just more organically and not a strict equation behind it.

And you can't say "Oh it's unfair to offload part of the algorithm to your brain, that doesn't count," because YOU DO TOO when you use the unsharp mask tool. Your brain is choosing each parameter using algorithms of its own, before the computer runs the USM algorithm. This is no different, just the brain is doing more of it.

I'm sorry, that is so silly as to be laughable.

There is no point in continuing a discussion... That article was just over the top on being absurd.

 

[Gavjenks]

The specific power equation is "gamma correction". Gamma is a measure of contrast. Look it up, and almost any dictionary will say it is "a measure of the contrast reproduced in a photographic or television image"

Look it up? Okay:
Glossary: Gamma: Digital Photography Review <--not an absolute measure, only a relative parameter in a power law equation
Gamma correction - Wikipedia, the free encyclopedia <--not an absolute measure, only a relative parameter in a power law equation
Gamma FAQ - Frequently Asked Questions about Gamma <--not an absolute measure, only a relative parameter in a power law equation
Gamma is undefined for anything other than a curve that fits a power function, which my random wobbly hand-made polynomial curve is most definitely not an example of.

"The key concept with curves is that you can never add contrast in one tonal region without also decreasing it in another. In other words, the curves tool only redistributes contrast."

Okay, that's nice, so what? I didn't claim to do anything other than redistribute contrast.
I fully admit: I robbed contrast from the areas of the image that didn't have very many edges in them, and give that contrast to the areas that DID have edges in them. Otherwise known as sharpening. Your beloved unsharp mask also conserves contrast, and does the same exact thing, just a bunch of times with weighted outputs. Robbing contrast from non-edged and giving it to edges is what sharpening IS. The blurring part of USM is just telling it where to do sharpening. The actual sharpening works the same way in principle.

Apparently I need an even more dramatic example.



Just wanna be crystal clear now... your claim is that the middle line is in fact blurry, and the line on the right is sharp?
Or are they both about equally sharp?
If so, I am interested in what your name is for the process that brings something from blurry to sharp, yet is not "sharpening."

I'm sorry, that is so silly as to be laughable.

Yes, you're right I'm sorry. Doing math and performing algorithms without an electronic computer is crazy and unrealistic, and there is absolutely no precedent for it at any time during human history. I am ashamed to have brought it up.

Also, I want to come clean on something else... This thing is a totally a hoax, man. Some historians and I made it up in the mid 90s just to screw with people. I sincerely apologize:

(wikimedia commons)

 

[EIngerson]

Sharpen.

 

[The_Traveler]

This is sort of like watching three females argue about whose penis would be bigger if they had one.

 

[manaheim]

*spits out coffee*

 

[Rick58]

This is sort of like watching three females argue about whose penis would be bigger if they had one.