[Aavso-photometry] Need Advice, CCD photometry;

[Michael Newberry]

>>  After you guys mentioned sigma clip processing, I went on a search to
>> find 'better' algorithims and I'm quite satisfied with what I've found.
>> Infact, I have chosen an alogorithm entitled SD mask, which incorporates
>> the best of sigma clip, median, and average.  If you guys have any
>> suggestions on the prefferred settings for these algorithms, I'd love to
>> here them.
>>
>>  I've also discovered that when combining there's another cool algorithm
>> to use entitled bicubic resampling, do you guys have any experience with
>> this?  Do you also use SD mask or sigma clip when combining your images 
>> or
>> just for calibration frames?
>>

 First, about bi-cubic resampling, we've had that all Mira versions since
 1992. I don't know if that still makes it cool or not! :-o)  But you are
 right,
 it is normally the preferred algorithm in comparison with bi-linear and
 nearest neighbor methods.

 The "SD Mask" is just another name for a technique known as "MTM Sigma
 Clipping", which has been used in the image processing world for decades.
 The letters MTM mean "Modified Trimmed Mean". There are slight variations
 from one implementation to the next on the way the sigma clipping iteration
 is done. People outside of astronomy have used that method in our Mira MX
 software for years. A year ago we added the method to the Mira Pro feature
 list for astronomers to use. MTM Sigma is *not* a panacea that combines the
 best of all worlds for all uses. It will give you smooth combined images
 because it clips lots of deviant pixels, regardless of whether or not there
 are that many deviant pixels in the pack, as will various other algorithms.

 This whole approach of clipping this and clipping that gets into a black
 art, and you have to ask yourself how much processing is "over-processing"
 for doing scientific measurements. The goal of any clipping strategy is to
 remove "bad" pixels, not just make an image look smoother. Getting the
 obvious ones is easy because they're so obvious, but it is very hard to
 distinguish the low level ones from good data. Using various clipping
 strategies, we can produce a combined image in which we measure a standard
 deviation of 2.4 but the true standard deviation should be 3.8 based on
 theoretical calculations. The difference is because the clipping method
 trimmed away "deviant" pixels that were just the normal outliers of a 
random
 distribution. Whether or not the difference in standard deviation affects
 you depends on what you want to do with the images.

 I strongly suggest trying a variety of combining algorithms, including
 Min.Max clipping, keeping in mind that the result having the smallest
 standard deviation is "not" necessarily the best image for doing science.

 Michael Newberry.

> _______________________________________________
>
> Aavso-discussion mailing list
> Aavso-discussion at mira.aavso.org
> http://mira.aavso.org/mailman/listinfo/aavso-discussion
>