Understanding Integration rejection maps

[mmirot]

I just did an integration of some dark frame with sigma rejection. There are plenty of cosmic rays on these darks( Kodak used radioactive glass on my CCD coverslip. )
For the most part the cosmic rays were cleaned off just fine. 
However, the high sigma rejection map is an even gray.  I would think these artifacts would be easy to spot on the maps.
Perhaps, I don't understand the maps.


Also, I was wondering if this sigma rejection is a standard type or a possion?

The last rejection mode " ccd noise model" could use some explaination too.  H
ow is this better for calibration frames? Which ones dark, flat, bias ,all of these?

I know my read noise and gain accurately.
I don't understand the third imput.
 

Max

[Niall Saunders]

Hi Max,

Do you have AIP4WIN, and the associated book "The Handbook of Astronomical Imaaging" (HAIP) ?

If not, then you might want to seriously consider getting hold of them (they come as a 'package'). Even though you choose not to use the software, the book alone is worth the money. It would certainly help you get to grips with some of the fundamentals of CCD image acquisition, and some of the basic steps in image calibration and post-processing.

HTH,

[Philip de Louraille]

I second that input wrt The Handbook of Astronomical Image Processing. Excellent book. The software, while nice, is not intuitive and not as interactive as it should (was probably designed on a PC by people who had a lot of DOS experience! ;-p  )  plus it crashes with pointer errors here and there but the technical descriptions are outstanding.

[Juan Conejero]

Hi Max,

However, the high sigma rejection map is an even gray.  I would think these artifacts would be easy to spot on the maps.
Perhaps, I don't understand the maps.

Rejection maps represent the number of pixels that have been rejected from each pixel stack. If a map pixel is black, then no pixel was rejected. If a map pixel is white, all pixels were rejected. Gray values are proportional to rejected pixels (for example, a map pixel = 0.25 means that a 25% of the total pixels were rejected).

You can binarize rejection maps if you just want to know which pixels were rejected, irrespective of quantitative data. For example, the following PixelMath expression:

$T != 0

can be applied to a rejection map to set all rejected pixels to pure white.

Also, I was wondering if this sigma rejection is a standard type or a possion?

The sigma clipping algorithm does not assume a particular noise distribution. It is the classical algorithm based on variability with respect to the median of each pixel stack. Variability is evaluated as the standard deviation. The algorithm is iterative and stops when no more pixels are rejected.

Averaged sigma clipping does use a Poisson noise model.

The next implementation (in version 1.5.2) is more sophisticated. I am implementing a more powerful rejection strategy known as Windsorization. It will be applied (as an option) to both sigma clipping and averaged sigma clipping.

The last rejection mode " ccd noise model" could use some explaination too.  H
ow is this better for calibration frames? Which ones dark, flat, bias ,all of these?

To understand the CCD Noise Model and Averaged Sigma Clipping algorithms, refer to IRAF documentation for the "imcombine" task. For example, in the following page:

http://stsdas.stsci.edu/cgi-bin/gethelp.cgi?imcombine

search for REJECTION ALGORITHMS (approx. 2/3 of the page). The mentioned algorithms correspond to the "CCDCLIP" and "AVSIGCLIP" IRAF algorithms. There are differences in my implementation, but the basic algorithms are the same. What we call "scale noise" is also known as "sensitivity noise" in IRAF. Normally you set this value to zero, since this noise (which originates mainly from flat fielding) is in general unknown.

My implementation of sigma clipping and percentile clipping are more different from IRAF's versions, but again IRAF's information is valid to understand the basic mechanics of these algorithms.

A significant difference of my implementation with respect to IRAF is that I always use the median as the reference value of each pixel stack, and never the mean.