ImageIntegration and master dark

[Simon Hicks]

Hi Niall,

I was lazily reading this post and then saw my name mentioned.... shock, horror....did I reeally say that? 

My point regarding synthetic bias frames was really to do with capturing the 'bias signal'. And this can be investigated quite easily (apart from all the caveates!) by simply looking at the bias frames that your particular camera gives. I know that there are different readout methods (though without any deep understanding) that give different bias signals. But if you get one of your bias frames and then stretch it to breaking point then you will have a better idea (at least visually....OK a gut feeling) for whether a simulated bias frame will do or not.

I think of a simulated bias frame as a smooth function.......the extreme case being a single value. If your stretched-to-death bias frame looks like it is a smooth function then job done. If however it has sharp streaks in it (mine does....Canon banding  ) or other sharp features then a synthetic bias frame will not reproduce this. And if you subtract a smooth feature from a sharp feature, you just create noise at the edges of the features.

So assuming you have sharp features in your bias signal....the next question is....are these somewhat reproducible from bias frame to bias frame. If they are then summing a few hundred of them will reduce the noise and leave the 'pure' bias signal.

However..........I've got this sneaky feeling that with the Canon cameras at least the bias streaks and sharp features are random from bias frame to bias frame, i.e. you can look at one bias frame and there's a clear set of distinct streaks....the next frame has a clear set of distinct streaks but in different places. Its like there's some background clock to the streaks, and a random readout timing....or something. So under these conditions, I am not sure that taking bias frames is much use at all.

But if your bias signal is somewhat reproducible...then yes, you should take bias frames, create a MasterBias frame and use it in the normal way......I think!

Cheers
          Simon

[Niall Saunders]

Yes Simon - a valid point that I had NOT taken into consideration !

I have only, to date, looked at the residual SNR of a MasterBiasOffset - and found that it was very low.

And that left me assuming that the 'spread of data' indicated by this low SNR was, obviously, very small. What I never took into consideration is that the 'spread' might actually still have an obvious 'pattern' to it.

Which makes me realise that data such as Darks and BiasOffsets really need to also have a Fourier Transform apllied - which should hopefully then show up any 'spatial' effects, such as the Canon 'banding issue'.

So, yes, unless I truly had no residual 'spatial' information in the BiasOffset component, then the application of a single-value SyntheticMasterBiasOffset would NOT be acceptable.

Thanks for making me think of this extra factor !

Cheers,

[Simon Hicks]

Plus, the word SyntheticMasterBiasOffset is far too long for polite company! 

[Niall Saunders]

Too long?

Well, it's not MY fault that the word 'Bias' just doesn't 'look right' in plural form - 'Biass', 'Biases', 'Biasses', 'Bias's', . . . ?

Hence I 'invented' the term 'BiasOffset' - easily made plural - and captures the fact that both 'Bias' and 'Offset' are terms that are equally as applicable to the data component that we are trying to refer to.

So, the natural progression is the term 'MasterBiasOffset'.

And, if you are going to create, or synthesise, your own frame, then why not .........

What, you don't think it will catch on  

Lexicographical Police - don't you just hate (detest, abhor, revile) them     (any excuse to use two smileys, one after another  )

Cheers,