Re: [APML]: Computer Enhancement

[Anders Thulin <Anders.X.Thulin@telia.se>]

Jerry Lodriguss writes:

> That is the whole point... to show through photographs what you can't see
> visually!

  This ties in well with the original concern that started this thread.
I'm not protesting against what you're saying -- just trying to illuminate
the problem.

  The question, then, is to know just when a computer enhanced image shows
ordinarily invisible reality (stretched greyscales are good examples) and
when they show visible non-reality.

  It takes a certain knowledge about image processing techniques to do that.
Just pressing a menu button is not enough, and no computer program manual goes
into any details. 

  Example:  In an earlier incarnation I worked with a company specializing
in image-processing hardware and software. The image-processing boffins
used a somewhat unusual image representation, and with much larger kernels that
ordinary binary operators use (5x5 was *small*, 13x13 or thereabouts was
standard). Combining these iteratively made it possible to do  (to me) almost
incomprehensible things.

  One of these was enhancing fingerprints, so that they were clean enough
to use for identification. In particular, these algorithms were very good
at teasing out good fingerprints from what appeared to be just a smudge.
Tests with experienced fingerprint experts demonstrated that the software
did just what they would have done in similar case. 

  Now, at one time this software was fed pure noise. Yup, the result was a
beautiful fingerprint. Enhanced into life by a computer.

  This doesn't demonstrate that computer enhancement is bad. You just have to
know when you can't use it, and the risks of using it on low-quality
data.

  Same thing with astrophotography. Doing a median filtering or any other
kind of interpolation too early may very well cause the 'real' signal to
get drowned in interpolated data later on.  

  You have to know what you're doing. And you also have to know when enough's
is enough. But if you do, computer enhancement is a great way to get
out all the light that hides in the dark spots of the image.

  For instructional purposes CE can be very useful. Just compare
David Malin's image of the Orion Nebula with anything that had been
printed earlier. 




Anders Thulin       Anders.X.Thulin@telia.se        013-23 55 32
Telia Engineering AB, Teknikringen 6, S-583 30 Linkoping, Sweden