To sharpen or not to sharpen

[apaflo]

The curves tool is one algorithm for sharpening. It's just a simple way to do one of the most fundamentally basic sharpening algorithms: contrast masking (rather than making an actual mask layer blah blah you can just mess with the curves and do the same thing, and if you actually use curved curves it works better than a basic mask anyway)

It affects some of the same variables as USM does, and the two influence each other in complementary ways, such that if you have curves + USM (with not all of its parameters), you can still achieve the same end result as if you had a better tool with more parameters given to you. Not perfectly down to the pixel identical, but almost / might as well be.

But a curves tool simply changes the tone map. It does not analyze surrounding pixels and make an adjustment based on what is there. It just looks at the value of one pixel at a time.

 

[Gavjenks]

So what? It ROUGHLY compensates for the things you were complaining about related to downrezzing (contrast and a slightly less surgical approximation of sigma at the end of the day). And it's something you always do anyway, so it's not even an extra step (just a slightly different curve).

I'm not suggesting that this is perfectly mathematically equivalent. I'm suggesting that if a person is really anal retentive about their sharpening, they can use this as a way of simply more closely approximating the look of sharpening after downrezzing, without having to fuss around with re-editing the stupid image every time they want to print it at a different size.

 

[apaflo]

So what? It ROUGHLY compensates for the things you were complaining about related to downrezzing (contrast and a slightly less surgical approximation of sigma at the end of the day). And it's something you always do anyway, so it's not even an extra step (just a slightly different curve).

But it doesn't! It has nothing to do with the sigma parameter used by a sharpen tool. There was no suggestion that downsizing changes the tone map. The curves tool is not something that should always be done anyway.

I'm not suggesting that this is perfectly mathematically equivalent. I'm suggesting that if a person is really anal retentive about their sharpening, they can use this as a way of simply more closely approximating the look of sharpening after downrezzing, without having to fuss around with re-editing the stupid image every time they want to print it at a different size.

Except that a curves tool does not sharpen an image. And if used to increase contrast between some set of tonal values, there is no way to avoid that it will also decrease contrast between another set of values. The curves tool just isn't related to the topic of sharpening.

 

[amolitor]

Just a little meat to keep you guys interested and going at it, because it's so fun to watch you wandering around in the weeds, sigma is a parameter for blurring, nor sharpening. It's the standard deviation of the ideal Gaussian used in a blurring step, which may or may not be part of a sharpening operation. Some sharpening operations use a blurring step, some do not.

 

[The_Traveler]

you two need to get a room.

 

[Gavjenks]

Curves does sharpen by itself. It does it slightly differently, but the end result is a sharper looking line, if so desired. Usually in addition to different contrast as well.
Normally that's an inefficient way of going about doing things. But if you're in a situation where the contrast and the sharpening both need changing in the right direction anyway (like this one), it works just fine to use a tool that changes both at once.


#2 and #4 are not the same edge. But they will look a lot more similar than #1 and #2. Hence "roughly" approximates, as I've been saying.
They will have different flavors, but they will all look sharper then the one on the far left. Call it strawberry sharpening and chocolate sharpening, whatever.

Also, in your brain, #2 and #4 will look more similar, because your brain already does unsharp masking to all edges. And it does it more aggressively when the gradient of an edge is sharper (such as #4 versus #1 or even #3 vs. #1), thus, the actual perception of Edges #2 and #4 will look more like this:


Which now suddenly look even more similar... it's sort of more along the lines of neopolitan vs. chocolate sharpening (chocolate is in everything!) The one that was already USMed will be stronger (and still possibly have a weirder, 2nd order shape to it now), but they will both have that characteristic shape in your perception, since the brain applies that to everything.

 

[tecboy]

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

 

[amolitor]

They both have incredible stamina.

 

[tecboy]

Just a little meat to keep you guys interested and going at it, because it's so fun to watch you wandering around in the weeds, sigma is a parameter for blurring, nor sharpening. It's the standard deviation of the ideal Gaussian used in a blurring step, which may or may not be part of a sharpening operation. Some sharpening operations use a blurring step, some do not.

I thought they were talking about Sigma Lenses.

 

[apaflo]

Just a little meat to keep you guys interested and going at it, because it's so fun to watch you wandering around in the weeds, sigma is a parameter for blurring, nor sharpening. It's the standard deviation of the ideal Gaussian used in a blurring step, which may or may not be part of a sharpening operation. Some sharpening operations use a blurring step, some do not.

Blur and Sharpen are essentially the same algorithm, with different values for the same set of parameters. Adjust them this way and it blurs the image, adjust them that way and it sharpens the image.

Unsharp Mask does use a blurring step. The high pass filtering of a Sharpen tool does not use a blurring step. The algorithm for a high pass sharpen has a sigma parameter, though the user might not be able to set it.

The sigma parameter is actually a parameter for a Gaussian Kernel generator and adjusts the Gaussian Curve used. Sigma is specifically the value for the distance from the center to a point at which the curve is 1/2 the peak value. Any algorithm that uses Convolution necessarily has a sigma parameter, even if the user does not specifically set it.

Hence every tool that uses convolution has a sigma parameter. Most sharpen tools use convolution, as do tools for blur and edge detection,

A couple cites:

Bluring and Sharpening -- IM v6 Examples

"Sharpen Arguments? (expand)

The most important factor is the sigma. As it is the real control of the sharpening operation."

High-Pass Filtering (Sharpening)

"High-Pass Filtering (Sharpening)

A high-pass filter can be used to make an image sharper. These filters emphasize fine details in the image - exactly the opposite of the low pass filter. High-pass filtering works in exactly the same way as low pass filtering. It just uses a different convolution kernel."

 

[amolitor]

 

[amolitor]

I actually understand what sigma is and how it is used in USM just fine, thanks.

You use words like 'Blur' and 'Sharpen' as if they were well defined and very specific algorithms, and they are not. They are just words. Probably in whatever tool you use, they mean something very specific, but we don't even know what tool or tools you use, and we certainly don't care. 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.

 

[amolitor]

The sigma parameter is actually a parameter for a Gaussian Kernel generator and adjusts the Gaussian Curve used. Sigma is specifically the value for the distance from the center to a point at which the curve is 1/2 the peak value. Any algorithm that uses Convolution necessarily has a sigma parameter, even if the user does not specifically set it.

Hence every tool that uses convolution has a sigma parameter. Most sharpen tools use convolution, as do tools for blur and edge detection,

This is complete nonsense. This so-called "definition" for sigma is not only wrong, but only makes sense for fairly specific convolution kernels.

 

[apaflo]

Curves does sharpen by itself.

Not by itself! It changes the tone map of the entire image, not just specific edge transitions.

You can sharpen using curves, but it is not particularly effect. First you have to use an edge detection tool to make an appropriate mask...

 

[Gavjenks]

I'll use that as an example of curves sharpening only. Looks terrible in this case / there's no real artistic goal of sharpening like this, but it demonstrates that it is quite possible. It looks terrible precisely because, as you say, it isn't masked for edges only, and there's a huge-ass section of the image with no edges AND midtones.


(Photocopier-chic)

But on a normal photo (as long as it doesn't contain a lot of sky OR the sky's tonal range is far away from the majority of midtone edges--which is most photos) it doesn't necessarily look terrible at all, because there usually aren't those huge midtone areas with no edges:


Notice contrast of course goes up, but so do the lines look sharper as well. Because all of those indistinct edges in the midtones are made steeper and thus much more visible. While edges in the highlights and shadows are mostly unaffected (very slightly dulled, but there are also way fewer of them here). And there's no photocopier-chic, because there aren't usually massive smooth gray gradients in real life taking up half the frame.

(Edit: more responsibly done with fewer inflection points)

It still technically will create semi-edges where there were none in smooth midtone gradient areas. But it doesn't matter, because in practice, all that does is just basically make leaves look a little rounder...




I almost always do this AND unsharp mask, and the two influence one another. Because this is IMO the most pleasing way to adjust contrast, but since it also sharpens a little bit (more of a side effect in my perception usually), you end up doing a bit less USM than you would otherwise as a result. The effect of each on the other doesn't require conscious thought, because you're just doing it by eye anyway, but it is happening. And I do always try to choose "peak" points in the histogram to center the curve on, for this reason.