Different kinds of sharpening

[greenmorpher]

Hello folks

I don't want to sound like a total idiot but I would be most grateful if someone could explain in words of not too many syllables when best to apply the three different types of sharpening offered by PL?

An example scenario or two would be great, and an indication of how to use the parameters.

I just don't grok it.

I fiddle around using the sharpeners and get some good results, but at other times, I get confused about what I am looking at -- because I try too many combinations.

Cheers, Geoff

[bruce1951]

Geoff I've been waiting like you for some sharpening education. Nothing.

But I'm not so sure it matters that much. Technically it may but visually I'm not so sure.

I only ever use Unsharp Masking. Then I use my eye balls. I zoom to the size the final image is meant to be and make the adjustments that please me. 'Please me' depends on my final use. I print A3+ for museums. At that size they need to be 'good'. BUT. That also depends. I have some images where I want the details and edges to 'pop'. So I deliberately over sharpen.

I'm sure that sharpening has a technically correct answer. But sharpening also has an 'artistic' affect as well. So my advise would be to go with what pleases you and wait until you get yelled at by the technically correct folks!!!

regards

[photoken]

... So my advise would be to go with what pleases you
...

Good advice. Zoom in to 200% on your image and try the different sharpening methods to see what works best. Apply the sharpening as an adjustment layer so you can refine it later and also add the other sharpening methods, if necessary. For example, I want to get the most sharpening possible for the RAW images from my compact camera, so the adjustment preset I created for them uses all three of the sharpening adjustments.

From a technical standpoint, I don't completely understand the differences between Unsharp Masking and Adaptive Sharpening, but I do know the difference between USM and Sharpening -- USM sharpens the highest contrast edges only, and Sharpening uniformly sharpens the entire image. That's why I suggest that the Automatic Correction filter should use Sharpening instead of USM -- cell phone photos need the sharpening applied to all image areas, even low contrast areas.

[bkh]

From a technical standpoint, I don't completely understand the difference between Unsharp Masking and Adaptive Sharpening, but I do know the difference between USM and Sharpening -- USM sharpens the highest contrast edges only, and Sharpening uniformly sharpens the entire image. That's why I suggest that the Automatic Correction filter should use Sharpening instead of USM -- cell phone photos need the sharpening applied to all image areas, even low contrast areas.

USM is the same as Sharpen if you set the threshold to zero, afaik. This means that Sharpen is probably faster than USM (but less flexible).

Applying a threshold makes sense for photos, however – ordinary sharpening also increases image noise. Usually, I start with a high intensity and medium threshold and lower the threshold until I can see increased noise (set the magnification to 100%).

As the manual says, USM applies the same threshold to each channel (e.g., R, G, B) while Adaptive Sharpen considers the colour difference (but there is no clear statement about how the colour difference is computed). This might produce fewer colour artefacts along sharpened edges. Also, Adaptive Sharpen seems to use a rectangular area to compute colour differences, and has a tendency towards artefacts when sizes become larger. It's probably the slowest of the three filters as well. I don't normally use it.

Btw., have a look at USM on the L or I channels only – generates less colour noise and fewer colour artefacts along sharpened edges.

Cheers

Burkhard.

[greenmorpher]

Good lord, and here I was thinking that I was some kind of special case in being confused by sharpening.

Thanks for these heartening responses!

Okay -- in all sharpening there is "Size". What is "size"?

in both Adaptive and Unsharp Masking, there is "threshold". What does that figure represent? A real measurement of some property or is it just like a notch on a twig so we know where we are and can move to a different position and make another notch?

Cheers, Geoff

[Hoogo]

I can add some technical stuff and a recipe how I use USM, maybe that's helpful?

[bruce1951]

Geoff my understanding is that PL's 'size' and 'threshold' are the same/similar to other editors. So any tutorial re sharpening should give you the appropriate information.

regards

[cathodeRay]

Geoff - I very much doubt you will be the first or the last to be confused about sharpening! As I mentioned elsewhere, I was wary of adding a comment lest I add to the general confusion. That said, here is my current and still very basic understanding:

(1) there is a distinction between so called input or capture sharpening and output sharpening (see for example the NIK filters). As I understand it, input sharpening is done early on, and is to make up for inherent blur in a digital image, be it from a camera or a scanner (not 100% sure about that last bit, as in how the device's output is inherently blurred) whereas output sharpening is done towards/right at the end of processing. Some add a third type of sharpening done in between the other two: creative sharpening, which is when you get to be creative (funk up the fur on that shaggy dog etc)...

(2) output sharpening is focused (sic) on what the output will be (screen/print etc). In the case of sharpening for screen, we have a very easy way of assessing what works: view the image at 100% (or more) and ask ourselves: do we like what we see? Print is a bit more complicated (and expensive...) (and others will be much more expert than me).

(3) as so often, terminology is as much a hindrance as a help. Unsharp Mask Sharpening? Hullo? Yes, I know its rooted in darkroom history, but... Deconvolution? De-what??? As I understand it, DeCo is a way of mathematically reconstructing an image, though I have yet to grasp the concept, let alone the maths. It might just be that if you can describe how the blur happened mathematically, then you can mathematically undo the blur...

(4) opportunity for further confusion abounds in other similar sounding filters eg NIKs ColorFx > Detail Extractor. I've no idea what the 'Detail Extractor' slider actually does, but the effect to my eye is sharpening.

(5) generally, sharpening means increasing local contrast, or edge contrast, as in making the gradient across an edge steeper (and that is also a clue to two common problems: overshoot, as it introducing artefacts/halos etc and aggravating noise, which I see as introducing random sharpening where it was no needed)

(6) there is often a case for applying a mask to a sharpening layer, and only applying the sharpening where it's needed, rather than doing a global sharpen. It is VERY easy to overdo sharpening. One can also adjust the opacity of a sharpened layer to control the global impact of the sharpening.

(7) sharpening creates/aggravates noise, so denoise before sharpening (and/or mask off relatively homogeneous areas eg sky before sharpening).

(7) as to the various methods:

(a) Sharpen (on it's own): like industrial strength acid in a domestic setting, it's bad news. Don't use it. If you do, expect deleterious results.

(b) USM: the 'general purpose' blunderbuss tool in the sharpening kit. The voodoo sliders are: size (or radius) controls the amount of blur in the blur ('unsharp') part and so how wide (in pixels) the effect is, Intensity (amount) which controls how much edge contrast is applied (of size is about how wide, intensity is about how high (steep) the gradient is) and threshold which protects lower contrast areas ie the sharpening is only applied above the threshold. I'm pretty sure the threshold slider is actually controlling a (contrast?) mask. The main problem with USM masking is you never get to see the unsharp mask - you can only alter it behind the scenes using the sliders. Sometimes the HiRaLoAm (high radius (size in PL) low amount (Intensity in PL) approach works well, sometimes it doesn't.

Since a picture is worth 1000 words, I attach a demo (16 bit so small to accommodate attachment size limits) which I find helps me understand what USM sharpening does (it also allows you to see what effect the sliders have). It's a two grey panel background with the border between the greys blurred (GB, 4px) and USN sharpened layer above, half masked so you can see before and after in real time. Zoom in to say 800% over the boundary between the greys and the masked/unmasked parts and play with the USM sliders. Hours of harmless fun...

USM Demo.pld

(c) High Pass sharpening (which I tend to prefer as I can better see what is going on): create a high pass filter and set the blend mode to overlay. The high pass filter finds edges (and you can see what it's found) and the Overlay blend does what it usually does ie darkens when the grey is darker than 50%/lightens when the grey is lighter than 50%. I tend to prefer High Pass sharpening over USM sharpening because (a) I can see what is going on (and can adjust and/or paint on the High Pass layer eg add some neutral grey to avoid sharpening areas I don't want to sharpen) and (b) it natively protects relatively uniform areas from random noise (they stay a neutral grey in the High Pass layer).

(d) Adaptive Sharpening: this term has a PL meaning (like USM sharpening, only differing in using all channels together instead of a channel by channels basis, ATTFM) but is also used elsewhere along with local contrast sharpening and other variants which increasingly use voodoo (ie black box methods). NIK's Output Sharpener has at least five sliders, four of which go from -100% to +100%, so you have almost an infinity of different settings you can use. They might as well call the sliders Voodoo1/Voodoo2 etc for all the meaning the labels have (nor does the he;p page shed much useful light eg at least teo sliders control 'adaptive sharpening'). You can also make the sharpening selective by using control points (physically determining where the sharpening applied. Frankly the only way I can use something like this is by playing with it, and the usual end result is a mess (unlike the Detail Enhancement in ColorFx, which I find does often improve an image).

So for me the bottom line is:

For general quick and dirty sharpening: USM sharpening (with or without HiRaLoAm settings)

For finer grained more controllable sharpening: High Pass Sharpening (I use this the most on photos that matter)

'Creative' Sharpening: either ColorFX's Detail Enhancement (or one of the other Contrast Effects: eg Pro Contrast, Tonal Contrast). Although these aren't called Sharpeners, given that the essence of sharpening is increasing local contrast, I am quite happy to consider these adjustments as forms of sharpening. I hope that doesn't upset too many experts!

Really desperate, worth trying anything (including the Nuclear option): have a go in NIK's Output Sharpener. Very easy to over-egg the pudding...

Hope I have managed to add more light than I have taken away, and that other's will kindly correct me where I have said something that is simply just plain wrong.

cathodeRay

Edit: corrected paragraph lettering under section (7)

[photoken]

Ray,
I strongly disagree with your (7)(a) summary of the Sharpen filter. I've found it useful as an automatic sharpening filter for images taken with my cell phone:

Sharpening comparison.jpg

Because it's effect is also applied to the low-contrast areas, it brings out the sense of texture in the sides of the buildings in this example, which improves the overall appearance of "sharpness". This is why I recommend changing PL's Automatic Correction to use the Sharpening filter instead of the USM.

I also use the HighPass filter as you described to increase the apparent sharpness. We've discussed this technique here in other threads, but this is my preferred way of achieving what other image editors call "Clarity". I use this as the last step (layer) when processing images, often after using one or more of the other sharpening methods.

[bkh]

Sharpening comparison.jpg

Is this really Sharpen vs USM with threshold 0? So far, I've been unable to see any difference between the two. (Since sharpening in PL's Correction Wizard doesn't have a threshold setting, I guess it's really "Sharpen", not PL's version of USM. But I haven't done any experiments on this, might as well be that it"s USM with a fixed threshold.)

Btw., The original (analog) USM filter doesn't have a threshold, either, so I'm not sure which should be labelled USM in the first place.

Cheers

Burkhard.

[photoken]

Burkhard,
As far as I remember, I simply applied the both the USM and the Sharpening with their default settings.

[photoken]

As the manual says, USM applies the same threshold to each channel (e.g., R, G, B) while Adaptive Sharpen considers the colour difference (but there is no clear statement about how the colour difference is computed).

This is where I get confused.

The help for Adaptive Sharpening seems to imply that the USM method is using individual masks for each colour channel.

So, my questions are:

1. Does Adaptive Sharpening analyze the combined all colour channels to create its single mask for targeting the edges?
2. Does USM first create a custom mask for each colour channel to target the edges, perform the "sharpening" on each individual colour channel, and then combine those custom "sharpening" results into one overall effect?

Yes, I've also noticed artifacts when "sharpening", but not so much of an increased noise. Although, to be honest, I'm so used to seeing the film grain from my days in analog photography that I'm not overly sensitive to what's called "noise"....

[bkh]

[*]Does Adaptive Sharpening analyze the combined all colour channels to create its single mask for targeting the edges?

That's what the manual says. (Technically, I don't think that PL creates a mask, but the decision is made pixel-by-pixel.)

[*]Does USM first create a custom mask for each colour channel to target the edges, perform the "sharpening" on each individual colour channel, and then combine those custom "sharpening" results into one overall effect?

Yes, USM works on each colour channel separately.

Really, USM (without threshold) is very simple. It computes a blurred image (Gaussian blur with the same size setting) and increases the difference between blurred value and pixel value. (USM is the same as a high pass layer with 50% intensity in Linear Light mode.)

If there's a threshold, the difference is only increased if it's above the threshold (thus the "mask" consists of all pixels in the high pass layer which differ from neutral grey by a certain amount, I guess).

Cheers

Burkhard.

[cathodeRay]

Ken - I've corrected the paragraph lettering under section (7) above but I don't think that affects your comment which is still about 7(a).

I've added two more adjustment layers to USM Demo.pld (and called it sharpen demo.pld): one for 'Sharpen' and one for Adaptive Sharpen' so all three can be compared (not perfect, but you can toggle between one and another). I'd hazard a guess based on what i can see that 'Sharpen' is USM without the threshold slider. Interestingly all three have Colour Model selection options, which rather seems to go against what the manual says about USM vs Adaptive Sharpen. Adaptive sharpen, at least at high intensity settings, is the harshest, and even has some over-shoot, so adaptive, but not necessarily in a helpful way?

Sharpen Demo.pld

What my previous post didn't cover was initial settings. For my normal preferred method, High Pass, you don't really need any because you get real time visual feedback about how the High Pass layer is shaping up. I just twiddle the sliders until the High Pass layer picks up just enough detail and then tweak from there. For the others especially USM it's really a telephone numbers game. My demo image (in which the initial blur on the background was a 4px Guassian Blur) can give some idea of how moving the sliders affects the image but in practice, for USM sharpening, PLs defaults are a good enough place to start (Size 1.5, Intensity 100% but with Threshold set to 0) and the general principle seems to be the softer the subject, the lower the Size/Intensity settings, and the harsher the subject the higher the settings. The Threshold slider is used to protect areas that don't need sharpening, and quid pro quo allows higher Size and Intensity settings. Sometimes (just to make things even more complex) it is suggested that Size and Intensity are at least partly complimentary, as in a higher Size merits a lower Intensity, and vice versa.

HiRaLoAm sharpening: start with say Size 30px and Intensity 30% and go from there (I use 30/30 as it's more memorable, some might say start with 30/50).

I have to say I curse the marketing goons who came up with such a plethora of 'appealing' terms. Sharpen? Detail Enhancement? Clarity? Adaptive Sharpen? Local Contrast Enhancement? Tonal Contrast? Pro Contrast? Don't they all do the same thing: increase local contrast, and what matters (to make the best use of them) is some understanding of how they go about achieving that aim?

There is a principle that applies in some fields (but not all) that when there is more than one way to achieve an end, it usually means none of them work very well (otherwise everyone would use the method that does work). Here 'work very well' includes the notion of usability, and I think that is a large part of the problem: all of the methods can sharpen, but equally none of them are intuitively usable. Too many complicate matters further by using fancy names and black box (voodoo) methods. As I mentioned earlier, this can be seen in full play in the NIK collection where - at least to me, even after I have read what 'manual' there is - it is impossible to know what most sliders are actually doing, except in so far as changes are visible in the image. Whilst this is good in that it is result focused, it is also haphazard and inefficient in that it is really just about applying random effects until something good comes out of it. More often than not, at least in my hands, the result ends up a mess and it's time to press reset...

[greenmorpher]

As you say, Ray, hours of harmless entertainment! LOL.

And rthat's what the High Pass filter is all about? Well! After a couple of experiments, I can see I am going to drive all my friends (and myself) bonkers by High Passing everything, particularly portraits of those with spotty skin.

A whole new world of fun is opening up for me.

Many thanks for the various explanations, folks. It really has been enormously helpful and will improve my pictures more than somewhat.

I will continue to browse through this thread and experiment now that I have a notion of where I am going.

Cheers, Geoff