Adobe Photoshop Training with Dan Margulis

Date: Wed, 6 Oct 2004 09:43:54 +0300

Hi - I have just come across a sharpening method using the high pass filter on a layer set to either hard or soft light. It seems to work very well on a lot of my images, at least on screen. It also seems pretty easy to get good looking results without accentuating noise, compared to using USM.

I find that setting the High Pass radius to 10 and then adjusting the layer settings improves most images.

I notice that there's no mention of this method in Dan's book - and would be interested to hear what people think of it. Any drawbacks or problems to watch out for? Also, while I know USM makes haloes, it is not clear to me what this method actually does to an image. I think if I knew what it was doing, I could adjust the settings more intelligently!

Most of my images get printed on a sheet fed press.

Toby Macklin

Co-create Ltd.

Nicosia

Cyprus

___________________________________________________________________________

Date: Wed, 6 Oct 2004 08:27:12 -0600

From: jim donovan

Hello Toby, I have had good success with the high pass filter. Try using "overlay" in place of "hard or soft light", I think it gives better results on some images using one layer. I find the high pass works well with fine detail,such as a field of grass or a lion's mane. Try "double hitting" the image with the high pass filter, 2 or more layers with different radius settings,one small,one large. Use one for hi-lites one for shadows,mabey different layer settings. I have found you can get drastic differences mixing it up with the high pass filter. Still prefer usm but I have gotten better results on a few images with high pass.Good luck, Jim Donovan

___________________________________________________________________________

Date: Wed, 06 Oct 2004 14:59:58 -0000

From: Graham.Scales

The high pass filter is good for "getting detail" out of a "flat" image by a change in contrast over all rather than appearing just round the edges.

Used carefully it can bring back shape to flat looking skintones on faces, accentuating the details such as cheekbones and bringing back some of the shading that may have been lost in brutal colour correction.

It's hard to explain without seeing it or being shown it.

It is also good as suggested to sharpen up images without effecting the noise or grain too much.

___________________________________________________________________________

Date: Wed, 06 Oct 2004 15:01:14 -0000

Hmmm...

It is not very interactive, but setting the blend mode first helps in some ways, but not others.It has no threshold. So it can be good to blur slightly first (0.2-0.3pix and or fades etc) and then HPS or blur/fade after when noise is a concern.

How many pixels are you playing with? What type of image content?

This is a lot higher radius than I generally use, but this is not to say that large halos are always a bad thing if done right.

As mentioned before, blur pre or post high pass/blend etc if noise is an issue.

Try various 'light' blend modes, hard light, soft light, overlay, vivid light etc. Try reducing opacity, try blend if slider settings or layer masks etc.

If in RGB, one can also desaturate the high passed layer.

One can also use all the other tricks as with USM with this method.

Also, while I know USM makes haloes, it is

Do you really wish to know. I spent a lot of time learning about the mechanics and application of USM - it was fun but there are better ways to spend ones time.

High Pass is different to other sharpening methods. From what I remember, it adds existing sharpness back to the image (amplified). The key to many regular sharpening methods is that they use a blur method behind the scenes (there is a low pass filter - these are terms also found in audio processing as well as image processing). It is the difference between the sharp original and the unsharp dupe used in processing which is used to make things sharper.

http://www.panix.com/~rbean/color/color5.txt

For a start, have a look at the DoG (difference of Gaussians) and DoM (Difference of Medians) actions in this link:

http: //members.ozemail.com.au/~binaryfx/Custom_Convolution_Actions_APS

5.zip

This is similar in net effect to high pass but not as sharp. The process is more similar to USM than it is to High Pass though (if I understand things right).

Totally different to USM.

In USM, one has the sharp original and a blurred copy. The difference between the sharp and the blurred is the halo or radius in USM. The strenght of the USM is similar to the blend mode/opacity which one blends the DoM/DoG/High Pass result with. USM has threshold which is not found in these other simpler filtering methods.

You can learn more about basic image processing via custom convolution kernels here (Java browsers req'd):

http://www.eas.asu.edu/~karam/2dconvolution/

http: //micro.magnet.fsu.edu/primer/java/digitalimaging/processing/conv

olutionkernels/index.html

More on USM and related methods:

http://www.photo.net/bboard/q-and-a-fetch-msg?msg_id= 001DGz

http://www.astropix.com/HTML/J_DIGIT/USM.HTM

http: //www.photo.net/bboard/q-and-a-fetch-msg.tcl?msg_id=000Qi5

http://www.photo.net/bboard/q-and-a-fetch-msg?msg_id= 001Mod

Best,

Stephen Marsh.

___________________________________________________________________________

Date: Wed, 06 Oct 2004 10:47:16 -0400

The high-pass filter finds (accentuates) edges. Hard/soft light shows areas darker and lighter than middle gray, so in this case it drops out pretty much everything except the edges.

An analogy would be to take a pen and ink and draw black lines around the edges of an object. Done very carefully, it would increase the illusion of sharpness. Here, radius corresponds to the width of the pen.

A disadvantage of this method is that it adds an image layer, so it increases the size of your file.

An advantage is that the effect is pretty much independent of magnification, so you can resize your image after sharpening.

I find it works well on some images, not so well on others.

___________________________________________________________________________

Date: Wed, 6 Oct 2004 11:31:26 -0700

From: "Paul D. DeRocco"

Unsharp Mask is a high-frequency boost filter. Except for the threshold, it will do essentially the same thing as mixing the image with a high-passed version of itself. The threshold is an added feature that prevents it from boosting noise.

You can accomplish pretty much the same set of effects, I think more easily, using Unsharp Mask, possibly followed by a Fade. For instance, Fade->Darken eliminates the white halos, but leaves the dark halos which are generally less objectionable.

Ciao, Paul D. DeRocco

___________________________________________________________________________

Date: Wed, 06 Oct 2004 10:55:36 -0700

From: J Walton

In my experience, there is nothing you can do with a single-layer High Pass that you cannot do better with USM. It is analogous to Levels and Curves, or Brightness/Contrast and Curves.

Dan has brought this out before. The reason High Pass 3seems2 to work better on some images is it tends to be a large radius and a low amount. If you find the High Pass with a radius of 10 works well, try using USM with a low amount (between 75 and 150) and the same radius. If you tweak the settings enough, you can get USM to look almost exactly like the High Pass filter.

This time, though, you are free to use Threshold and are free to adjust the intensity of the white/black lines. I encourage my operators to understand the USM filter rather than use High Pass. But, as always, Your Mileage May Vary.

J

___________________________________________________________________________

Date: Thu, 7 Oct 2004 16:26:56 +0300

Thanks everyone - lots of ideas. I am going to experiment, with the high pass filter and with different settings in USM.

On the high pass sharpening, could someone tell me how the 'Blend if' dialogue works? I understand that it will limit the areas that get sharpened (or whatever), but I'm not sure how to make it work. Do I need to look at each channel separately? What about the Gray channel? And I see that I can split the sliders. What is this doing?

Toby

___________________________________________________________________________

Date: Thu, 07 Oct 2004 15:45:15 -0000

I agree. Early on in image processing, I now tend to apply a USM with radius 60 (yep, 60), amount 10-15 and whatever threshold keeps the grain out of the way. It seems to boost local contrast in exactly the same way as high-pass/soft-light, but is more versatile.

Robert Johnston

___________________________________________________________________________

Date: Fri, 08 Oct 2004 01:20:16 -0000

Agreed Robert, but I would probably be more inclined to hit the image with a minor contrast mask blend or CS Shadow/Highlight first, before the marginal boost of contrast with the wide halo/low amount USM.

High radius low amount USM is a different goal than sharpening for digitization or for output reasons.

Stephen Marsh.

___________________________________________________________________________

Date: Mon, 11 Oct 2004 01:44:05 EDT

Stephen writes,

That's correct. As pointed out by J Walton and others, this method is better, because it's more flexible, than using the High Pass filter. As for comparing to the conventional method of sharpening with a high Amount and a low Radius, well, that method is conventional for a reason. Usually it works better, but there are exceptions. I discussed them in group message #4304, or one could consult the full thread,

http: //www.ledet.com/margulis/ACT_postings/ColorCorrection/ACT-LAB-sharpen.htm

Dan Margulis

___________________________________________________________________________

Date: Mon, 11 Oct 2004 15:28:32 -0000

Dan Margulis wrote:

Hi Dan. I don't use HP filtering for the sort of contrast kick that we are talking of with high radius and low amount type USM (say amount 5-10%, 50-80 radius, 0 threshold).

I agree USM is more interactive/flexible.

I like HP as it can pick out some image frequencies that USM does not. For some images this can add a little lift. It may not be applied globally though.

Leaving low amount, very high radius USM or large amount HP contrast blending moves and back to the general topic of the HP filter vs. USM...

There are fundamental differences in the mechanics of USM and HP sharpening. For those wishing to know more on the process:

The HP filter blocks lower image frequencies (lack of detail) and passes through a selectable range of higher frequencies (this is the slider in high pass).

HP filtering often enhances more noise than similar USM settings. Minor blur before HP filtering or afterward can help (the find edges filter picks out noise, while the glowing edges filter has a blur function built in on the fly...the HP filter could too!).

The result of this basic image processing is a neutral 50% gray for areas which are not high frequency and passed through the filter. This means that edges (high frequency areas) are defined by a halo, just like USM.

By using a blend mode that treats 50% grey as transparent - we add the edge definition to make things appear sharper. Blend modes such as overlay, soft/hard light, vivid light etc add contrast and amplify the effect of the new edge definition.

The end result is similar to USM with a lot more manual work. All things being equal, the two methods do provide subtly different results - as does DoG or other sharpening methods. For some images the slight differences may be preferred over another method.

USM is different, instead of passing through a certain range of image frequencies - the original and a blurred dupe are subtracted from each other to create the edge definition.

For most images I stick with USM.

But if I have the time to trial, then I will compare many different things - including using HP as an alternative, but more often as an addition to USM (if USM is the icing on the cake, a tweak of applied

HP filtering could be the cherry).

Stephen Marsh.

___________________________________________________________________________

Date: Mon, 11 Oct 2004 16:57:06 EDT

Do you have examples? Because when I was testing this last year I didn't find any cases that HP gave me anything I couldn't get with the USM filter a lot quicker and more flexibly. I wasn't trying to duplicate the files, of course, but am wondering if there is ever actually a significant quality difference.

Dan Margulis

___________________________________________________________________________

Date: Mon, 11 Oct 2004 21:36:47 -0400

An alternative to either the High Pass or USM Filtering method is to use a blur layer that works very similar to High Pass sharpening but is even more subtle.

I dupe the base layer and then apply a Gaussian Blur in an amount that is needed based on the image size and resolution (same as using the unsharp mask filter or high pass filter ). I then invert the blurred layer and set it's opacity to 50%. This in effect creates a high pass channel that can then be used for sharpening but is even more subtle in edge detection then High Pass Filtering or other edge detection filters.

To use it I simply add an empty curves correction layer between the original base and the blurred dupe and set the curve layers blending mode to overlay or hard light. I then establish a clipping group between the curve layer and the blurred dupe by option clicking between the two layers. (The curves correction layer is merely a holding place to use a blending mode and adds nothing to the file size.)

This method works very much like High Pass Sharpening but will be even less offensive to grain than either High Pass or USM. I frequently use it on low quality digital captures that have been supplied for high end use and could stand a little sharpening where other methods cause too much damage.

With a little finesse you can even restrict this sharpening to just the luminosity or even split the darkening or lightening effects and control them separately.

Regards,

Jerry

___________________________________________________________________________

Date: Wed, 13 Oct 2004 10:34:09 -0000

I was talking of that little bit of extra 'sparkle'...the cherry on the icing. HP does find slightly different detail than USM due to the mechanics behind the filters.

Please download the file found in the Yahoo Files section for this group, named: USM_HP.zip

This has three files.

The original so that you can test and evaluate for yourself.

Two example files.

One is a difference blend of the high pass filter at the min effective setting (0.2 rad.) from the original. This has no further amplification of effect as would happen when using a blend mode such as soft light, overlay etc.

The other is a difference blend of the USM filter at 500%, 0.2 rad. and 0 threshold. This has been amplified via the equalize command to make the effect stronger and more visually apparent.

One can see that HP picks out more minute detail than USM - even when USM is amplified it does not pick out the same detail.

This is the sort of subtle difference that I was remarking on, which for some images one method may give a more 'artistic' or 'more visually pleasing' result.

Stephen Marsh.

___________________________________________________________________________

Date: Fri, 15 Oct 2004 17:15:06 EDT

Stephen writes,

What I see is that the results of the two are different. Considering the black-and-white linework used to illustrate the effect, it's hard to visualize that there might be a real-world impact--it's the type of image we would never sharpen, done with a Radius we would never use, followed by a contrast enhancement that we would never use either, all without the benefit of a Threshold that we usually put it.

While I'm certainly open to looking at an image that shows a perceptible advantage for doing it with High Pass as opposed to a carefully chosen USM with a similar Radius, I've run a few more tests and I am very unconvinced. It would seem that the only time it couldn't be duplicated closely is specifically when Threshold=0 and Radius <3, both of which are uncommon settings in the sorts of images that are amenable to this type of sharpening. Even then, the only t

ime it would make a difference is when one wanted *both* the major halos and the ambient noise to be exactly what they were using High Pass.

There has to be some reasonable balance between trying to squeeze the most out of the image and the effort it takes to do so. A nearly infinite number of valid sharpening possibilities exist, but many of them offer minimal gains for maximal work. One of the most troubling areas for the next edition of Professional Photoshop is how to expand the sharpening section without making it several hundred pages long.

The first in a series of three articles exploring how the blur affects the sharpen has just appeared, in the context of blown highlights/plugged shadow. The second shows how the concept works in the Shadow/Highlight command. And the third tries to sum it up, opening with a demonstration that one can sharpen extremely effectively using only the Gaussian Blur filter plus layer blending modes, in a manner somewhat similar to the technique Jerry describes but with more nuancing available. The result was great--but it took 23 steps. No way that's worth it in the real world.

Dan Margulis

___________________________________________________________________________

Date: Fri, 15 Oct 2004 18:51:32 -0400

It might, as an action.

Laurentiu Todie

___________________________________________________________________________

\

Date: Sat, 16 Oct 2004 21:24:27 -0400

Hi Dan,

I tend to agree. Two years ago I was experimenting with a gaussian blur/ sharpening methodology in an attempt to reproduce the sharpening results of older analog drum scanners. I was able to come very close but the number of steps involved was way to much for real world working conditions and not very repeatable. I did reduce it to an action, as Laurentiu Todie suggested, which included stops for decision making on the fly, but it still took nearly ten minutes to process a medium sized image. Not practical when you have a few hundred images to prepare and sharpen. The method that I had described in my previous post was a very simplified example and works well on low quality digital captures that are either very grainy or have severe artifacting or both, but still require some sharpening for offset printing.

I look forward to reading your articles.

Jerry

___________________________________________________________________________