Weird Noise

[RichLD]

On 08/11/2018 at 09:05, Whirlwind said:

I think you might be misunderstanding what dithering is there to do.

At a basic level there are two types of noise that need to be considered. 

1) There is a type of fixed noise which may be better described as known errors on the camera.  These are things like hot pixels, cold pixels and so forth, bad columns and so forth.  These stay in the same location all the time.  Darks and bias can partially correct for these but they can accrue charge in a non-linear way which means even with dark/bias correction they can still partially exist (this is particularly relevant for non temperature controlled cameras).  If your tracking is so accurate that the points will always fall on the same point in your image then when you combine them it will reinforce this artefact and you will end up with little bright or dark spots over the image.

2)  There is random noise.  This comes from that it is statistically impossible to measure something at the highest levels of precision.  If as an example the flux arriving at a pixel is 300 counts then when you image the object you would find that the you get a range of values with greater frequency towards the actual value.  So over five images you may get the following values 295, 298,299, 301, 308.  This is called random or gaussian noise.  It is impossible to predict the exact value you will measure from image to image.  However, it is random and follows a gaussian statistical variation.  As such because some values will be lower and some higher you can median or mean (average) combine such data and the result will trend to the 'real' value of 300.  The more images you take the more you can average out these random fluctuations.  Hence this is why you take multiple images and stack them.  When you process the data you exaggerate this effect as you are trying to pull out the slight variations to the signal (for example in a nebula) that might be very similar to the level of random noise you get.  Hence with too few images averaged then what is real and what is noise becomes confused.  This is what causes the mottling effect.  The more images you take the more you can average out the background noise and the more certain you are as to what is data and what isn't.  Hence when people refer to images being overdone, overcooked this is what is happening.  The details have been over processed to the point that the noise is being processed not 'real' signal. 

So why dithering?  Well dithering is a random jump of your camera close to the target of interest.  It makes no difference to random noise because it is not based on a specific location.  Wherever you point your camera you will get random noise.  On the other hand the fixed point noise (like hot pixels) are tied to specific pixels, columns and so forth.  As such they will move about your image when your slightly shift (dither) the telescope.  Most software when it combines data will reject any that are hugely discrepant from the average data in your image when tied to a fixed position through star alignment.  So suppose you had a hot pixel that had been dithered once at a specific sky position.  You may get something in counts like this:- 300, 299, 306, 295, 301, 2000.  The software will recognise the last value as discrepant and then ignore that data for that specific pixel and reject it.  If you don't dither you would get something like:- 1999, 1998, 2003, 2007 2003, 2000.  Hence the software won't recognise that this is a hot pixel but consider it a real value and include it on your image.  What you end up with is an image that has scattered single pixel bright spots because of this type of fixed noise.

As such when noted above that your image looks like dithering is working it is because there doesn't appear to be any single pixels that are overly bright for their location.  The noise you are looking at is random noise which can't be fixed by dithering, only by taking more images and you can never get rid of it completely.  So it becomes a balance as to how much time you want to spend on an object and how much processing you can get away with before the remaining noise becomes distracting.

 

Brilliantly explained!