: Photoshop: high pass filter or sharpening filter? I was wondering, what is the difference between the high-pass filter and the unsharpen mask one? I am used to use the first to improve the quality

@Welton168

Let me tell you a secret, Photoshop is doing almost the same in those cases.

Let me explain.
Classic sharpening (Using Linear Filters) is as following:


Take an Image.
Apply High Pass Filter on it.
Add the result (Or some kind of adding operation) of 2 to the original image of 1.


More than that, in Photoshop the High Pass Filter used in Unsharp Mask is the exact one in the High Pass filter.
Actually, in Photoshop, both are created as following - Photoshop's High Pass:


Take an Image.
Blur it using Gaussian Blur (Which is actually Box Blur applied few times).
Subtract the result of 2 from 1 and put the result as High Pass Filter result.


Now, the question is, given an High Pass of an image, how do you want to use it?
If you just want to add it, then Unsharp Mask is your friend.
If you want different application (Like Multiplication and addition, etc...) use High Pass Filter + Calculation Tool / Apply Image / Blending Mode.

If you really want a modern way to sharpen images - Use Multi Scale / Multi Frequency Sharpening.
Those are superior to any of those.

Just Google Multi Scale Sharpening Photoshop.

Another great approach to deal with the weaknesses of High Pass based Sharpening (USM included) is given by DoubleUSM.

Davide Barranca is an expert on this and he has amazing and hilarious video on the subject - Double USM 2 Panel for Adobe Photoshop.
In summary, he take care of Dark and Bright Halos in an independent approach, tackling each on its own.

Vote Up Vote Down

@Jamie315

Generally, low-pass filters blur, high-pass filters sharpen.

Unsharp Mask is BOTH a high- and low-pass filter.

So you:


Make a low-pass filter (radius parameter) - Gaussian blur - result in a low-frequency image
subtract this from original image which results in a high-pass image this is the unsharp mask
Make a ramped contrast image of original (increased contrast is a sort of high-pass operation)
Compare luminosity of the above two images (mask vs contrasty): 100%? then use contrasty pixel; 0% use original pixel; 0 < luminosity < 100 ? blend


To sum up: the sharpening is a high-pass operation blended with the original using a weighting process influenced by a low-pass filter.

Vote Up Vote Down

@Gail6891361

Unsharp mask* is a highpass filtering technique. To be more precise Unsharp mask is original image plus a highpass filter. You can use a highpass layer to accomplish the same thing, yes.

You can also use highpass filtering for other things, such as frequency separation, masking of noise, fog removal etc... So unsharp mask is just one implementation of highpass filtering.

* if we wanted to be correct we should call it unsharp sharp mask. But that kimd of name sounds tedious so we do not call it that.

A more elborate explanation

High pass filter in Photohop is implemented via the same low cut** filtering as Unsharp mask. Thus one can say that they are in effect effectively the same operation family. The high pass is a bit optimized for Photoshop but level workflow so the effect is offset and you do not end up with negative values. And operating via internals of blur certain steps can be omitted.

There is a old technique used in photolab that indeed was a single unsharp mask. However due to mathematical decomposition of this you end up with more or less the same thing as copy of image screened with what Photoshop calls high pass.

Unlike the photolab version though the computerized unsharp mask filter has a clear sharpening stage. The photolab version also may have this stage but its more implied. Therefore the naming is slightly misleading, but works well for people who come from a older workflow.

This can be easily verified by building the layers up in photoshop from a blur. If one neglects the small errors that comes from quantisation and rounding in extra operations.

** Low cut is same thing as high pass. At least theoretically. In some cases implementation details may vary a bit for this to be untrue, the aim is the same. In this case the difference is nonexistent.

Vote Up Vote Down