Dark and Bias Values on STF8300

[Oscar in Bin]

When I take a Dark Frame I would expect the mean ADU to be greater than that I measure on the Bias given the Dark includes the Bias. This is not the case. So when I subtract the Bias from the dark I end up with a negative value.
This all seems rather weird and I hate to think what it does later in the calibrations.
Thoughts? :confused2:

[bert]

Same temperatures?

[Oscar in Bin]

Yes. Both taken at minus 20

[cventer]

How many bias and dark frames did you take ?

You need at least 12 - 20 bias frames combined with median combine

And at least 12 - 20 darks combined with min max reject algorithym before any meaningfull analysis.

[gregbradley]

What do you need bias frames for?

Are you making adaptive darks?

Greg.

[Oscar in Bin]

Greg, using them to scale the darks. However, because of this problem I am now matching dark exposure times to the actual imaging times.

In answer to the question about the number of frames I was taking 40 and doing a mean after a min/max reject of 2 each so data should have been pretty good.

[gregbradley]

Sounds good. I thought sigma reject was hard to beat but I suppose you are doing in 2 steps what sigma reject does in one. Both approaches get rid of outliers that are artifacts rather than repeating noise.

40 sounds a lot. I think I could probably upgrade my darks as my biggest dark is about 16 and many are only about 8. But then my FLI cameras have practically no noise so its far less important than for other cameras. Still I find even with such low noise my CDK17 has quite a hot spot due to its design and corrector. Without a decent dark I find the flats just don't work. Strange. Anyone know why that would be?

Greg.

[Paul Haese]

I'm with you Greg withh using Sigma reject and 16 darks, but I do use 16 biases too.

[gregbradley]

Yes mean is the average of a sequence. So its kind of a pretty rough approximation of what your average pixel should be. If you have lots of hot pixels it will throw the calculation off quite a bit. I suppose that is why you run min/max first so the mean becomes more accurate.

But min/max probably only throws out values above/below a certain range. Again that is crude as any image has bright and dark areas and they are not artifacts.

Sigma reject is more complex computation that is pinpointing more accurately the pixels that are quite clearly an artifact and outside what is statistically expected of pixel values taking into account the whole image.

An ideal computation would detect every artifact yet leave intact bright and dim areas without change. That is the problem the mathematicians are trying to solve with these various combine methods. It is my understanding (please correct me if I am wrong) that sigma reject does this the best of the current methods. Although I see CCDstack defaults to min/max. I should look into how that is calculated.


Greg.

[naskies]

While I can't speak to the specific methods used in astro software, I do have a basic understanding of mathematical statistics so I can make some theoretical guesses...



The raw image is effectively divided by the flat image on the individual pixel level (i.e. multiplying the dark/vignetted parts so that they are brighter), so without a dark frame you won't be subtracting away the bias and dark signals from your image first. This means that you're increasing the noise of the bias and dark signals - along with the true signal - when attempting to correct with flats.

With a dark frame, the bias and dark signals are subtracted - i.e. effectively become zero - so that only the true signal is being increased.



I'm a bit surprised that compressed sensing (http://en.wikipedia.org/wiki/Compressed_sensing) methods haven't been used for image reduction yet (to my knowledge, anyway). That stuff is like voodoo magic - it can re-generate images from less than the theoretical minimum sampling than the Nyquist theorem predicts.

(It works because pixels in real world images aren't statistically random like we tend to think - adjacent pixels are actually highly correlated.)