From: Chris Russ <jcr6@aol.com>
Subject: Re: Unsharp masking
Date: 25 Oct 1999 00:00:00 GMT
Message-ID: <3814CCA6.A1194762@aol.com>
Content-Transfer-Encoding: 7bit
References: <940558311.102016@jaguarundi> <940636211.24257.0.nnrp-11.9e9810ec@news.demon.co.uk> <7utlnr$ie4$1@the-fly.zip.com.au> <38127660.117E1A70@NOSPAMeecs.umich.edu> <7uu1f4$bh3$1@the-fly.zip.com.au> <3814697E.22B4057B@aol.com> <3814B121.EC7442B4@pandora.be>
X-Accept-Language: en
Content-Type: text/plain; charset=us-ascii; x-mac-type="54455854"; x-mac-creator="4D4F5353"
X-Complaints-To: abuse@prserv.net
X-Trace: 25 Oct 1999 21:33:41 GMT, 32.100.213.180
Organization: Reindeer Games, Inc.
MIME-Version: 1.0
Reply-To: jcr6@aol.com
Newsgroups: sci.image.processing,comp.graphics.algorithms

Martin Brown wrote:

> The question is whether or not that is what unsharp masking really means in high
> dynamic range images. If it were that simple the results would be no different to
> high pass linear filtering.
>
> The orginators of the technique for scientific imaging at AAO have a web page
> which contains some technical details and the implication is that the final image
> is the result of combining in a *multiplicative* fashion the original image with
> a blurred negative copy.

You could always divide by the Gaussian Blur instead of subtract... That will
make a difference.  (This comes dangerously close to bringing in the whole
discussion of Gamma, but since we're talking about enhancing high frequencies,
the goal is to make things perceptually sharper, not keep exact intensities.
So, let's not.)

There are really two common uses for the Unsharp Mask:

1) Reduce the dynamic range of the image so that a much broader range of
initial intensities are now visible.  (Especially conversion from Film or Plates
down to prints.)

2) Slightly enhance high-frequencies and possibly clip a little at black & white.
(Sharpen after some kind of blur operation has reduced the noise.  More like
an HF Boost.)

Remember that even though an Unsharp Mask operation is technically a
high-pass filter, the roll-off on the Gaussian is so slow that you have to
get a pretty large one before you start really cutting off much.  Thus,
for small radii, the USM is more of a High-Frequency Amplifier.

Photoshop's Threshold operation was not possible back in the "analog"
days of film, but is very nice for certain kinds of operations.  If you don't
like it, set it to 0.

re: -1 -1 -1, -1 3 -1, -1 -1 -1 kernel

> This is still a linear filter and not what I would understand as unsharp masking.

Without the threshold operation, it *IS* a linear filter.  So is a High Pass.

-Chris Russ