View Single Post
  #5  
Old November 23rd 07, 06:08 AM posted to alt.binaries.pictures.astro
warner[_4_]
external usenet poster
 
Posts: 4
Default ASTRO: Holmes on the 18th UT reprocessed



Rick Johnson wrote:

Working from JPG images tends to result in just enhancing the artifacts
of JPG compression. This is why you see all the point like sources in
the middle of the three images in this series. Also the FITs image
contains a brightness range of 100 to 62000 ADU counts. Only 256 of
those fit in a JPG image so most of the data has been thrown away before
you even get started. But its the JPG artifacts that really can lead
the processing down the wrong path.

Also it is best to remove the color data before doing any processing as
that too contains artifacts adding to the problem.

Given the tremendous amount of data in a FITS image compared to what a
computer screen can display you can play this enhancement game for years
and never see it the same way twice.

When pushing even FITS data hard you have to look out for artifacts
forming. They are inevitable and you just have to live with them and
recognize which features are real and which are artifacts.

For instance when doing a heavy unsharp mask you will get ringing around
a bright edge. But only on one side. Since a bright star has only one
edge (the outside) you get an intense darkening surrounding the star
often called Panda Eyes. But with the bow shock of this comet being in
an arc as shown in another posting here you get a darkening mostly on
the side of highest contrast, again the outside so you see a strong
black arc you labeled as a void in that picture. It's just the type of
Panda Eye you get from such a feature. Without seeing the original data
I can't tell but suspect that many other of the voids are due to the
mathematics of unsharp masking as well. Some may be real. Only looking
at the original unprocessed data can you tell which is which.

For the attached image I worked hard to give a strong unsharp mask but
still prevent these types of artifacts from developing. This takes many
many steps and takes me several hours of very slow progress. After
running for about 10 hours the attached is the result. The graininess in
the bright areas is mostly noise and not real -- I think. As I slept
through most of it and didn't watch its progress I can't say for sure
but with only 4 minutes of data noise seems the likely explanation. The
routines are applied with varying intensity depending on the signal
strength. It appears I applied them too weakly to faint signals and too
strongly to bright ones as the faint areas are too smooth and the bright
too rough. But I don't feel like spending another ten hours trying a
different curve so this will have to do.

No matter how I processed the FITS image I saw no sign of those point
sources in the middle of your last set of three images. I think they
were the result of how your processing handled the noise and JPG artifacts.

Back in the 70's I designed digital math circuits for processing audio
data for high speed teletype experiments the FCC let me and a few other
hams run. I had to develop math routines to deal with just this type of
artifact though in audio we called it "ringing" but in visual image
processing you see the "ringing" as Panda Eyes and other artifacts. The
bright rings around stars from deconvolution is another form of it for
instance. So I used routines to develop my own filters that I used on
this attached image to get a lot of unsharp masking with little ringing.
They work just fine visually as well as with audio but really tax even
today's computers compared to what was needed for the same at audio
frequencies. There, before even the earliest 4 bit processors, I used
TTL gates to do the processing in real time.

Rick


Rick I very much appreciate your comments here. A real education
one could get nowhere else. Frankly, I a not surprised at anything
you say; just the explication of it which it would take hours to find
elsewhere. This helps tremendously.

Very sincerely,
Jerry Warner





warner wrote:
more
Jerry



------------------------------------------------------------------------
[Image]