[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [TASS] MKIV Darks and Flats
On Fri, 17 Dec 1999 00:24:00 GMT, Andrew Bennett <andrew.bennett@NS.SYMPATICO.CA> wrote:
*>1) "Dark" is point-by point median of 2032x2032 image area for the 3
*>dark images for each camera on CD5. Not great, being median of
*>just 3 values per point, but it does catch enough of the hot
*>points to save Star from getting bamboozled. The first version
*>has no allowance for variation of the mean or scaling between images.
The darks don't vary a lot between their averages.
*>I intended to write a program to use a catalog of hot
*>locations and to replace every hot point with a median of
*>adjacent points. This program got put on hold when the simple
*>subtraction "worked" but I think this is the right way to go.
*>There are just too many intermittent hot pixels not to do
Intermittant dark pixels are called "noise", or "cosmic rays". There
are some CCD phenomena that involve trapped charges that may produce
variable effects: don't sweat them. That is one reason Richmond
used multiple observations in his database construction.
Rather than review the business of darks, hot pixels, and so on,
I emphatically recommend the following. There is an excellent book on all
this and more: Christian Buil's "CCD Astronomy", available from
Willmann-Bell. Andrew, buy this book, it is a well-known reference to
amateur CCD users.
*>2) "Flat" is a point by point median of the 35 image areas. Each
*>point is dark-subtracted and scaled by 32768/(image median -
*>dark median). This is pretty rough too; I didn't purge the
*>cloudy images first. Or the airplanes. Or those fascinating
*>bands, whatever they are that don't look like clouds or airplanes.
*>Applying these with my Borland Pascal 7.0 program, again without
*>any allowance for Dark scaling, reduced the noise level reported
*>by Star by almost a factor of 2. I keep meaning to try Dark and
*>Flat separately to see which does most for the noise but have not
*>got round to it.
As for dark scaling, if the value of a dark column in a dark is
close to the value of the dark column in the image, no scaling is
needed. "Close" is the operative word, I've not looked this over.
I could do this in half an hour if necessary.
When I look visually at some of the Disk #5 H3 images, they have what
looks like a cross of fog; some have blotches. These are not a lot
above the minimal background of around 12500 or so. Any non-repetative
background would be reduced by n/35; random noise is reduced by 1/root 35
or 1/6. My Astrometria reviews of the H3R1438.* images (raw, no darks, no
flats) gives the following numbers, which I presume are average background
and one standard deviation: a low of 11983 and 379 to a high of 12368 and
419. Those images are pretty "clear" compared to other images in the set
though I have not looked at all of them.
*>3) At least as important as Dark/Flat for getting Star to work
*>properly in catalog matching is getting rid of the "saturated"
I agree that attempting to measure saturated sources is at least
problemmatic. That's why I've been working on trying to find by
inspection the value of saturation. Tom can read this off the specification
sheets as one check - although those sheets can be wrong. I'd say
it's well above 40000 ADU's and somewhat below 50,000 ADU's, with BZERO at
32768. (i.e. add 32768 to 50000 for the values of likely saturation as
represented by 16-bit values in the FITS files). I'll Web-publish my data
as I catch up from my illness. But I'm doing it the hard way, by inspection:
the easy thing would be to accept what the spec sheets says but I'd hate
for the spec to UNDERestimate saturation.
Herbert R. Johnson http://pluto.njcc.com/~hjohnson
email@example.com voice 609-771-1503, New Jersey USA
amateur astronomer and astro-tour guide
S-100 computer restoration, parts, manuals as "Dr. S-100"
rebuilder/reseller of compact Macs for your computing pleasure