Combining Images

[roundycat]

Hi all,

having struggled to understand why some people choose to use the software/combination methods they do I thought I would put pen to paper and describe some tests that I have done. The link takes you to a summary of using Maxim and five different methods of stacking subs. The results did not surprise me as this is the second time I have done this. The only way you can really get to the bottom of some things is to go on the journey yourself. I would be pleased to hear any comments, both good and bad.

http://www.dens-astropics.org.uk/page%2019.htm

Dennis

[brendant]

Great article Dennis, with interesting results, thanks for sharing.

Brendan

[MartinB]

Thanks for that comparison Dennis. The jargon around combining and other astro imaging processes can be pretty daunting

[Psychobilly]

Worthy of a Sticky - I think so.... as people are often confused as the benefits of the various combining methods...

Billy....

[roundycat]

Here are some more results from an attempt to find out what SD Mask is doing with noise when combining frames.

I made a set of 250,000 pixel frames in Maxim and added Gaussian noise giving pixel values from 0 to 126. This is the nearest I could get to an ordinary light frame with a smooth area with nothing but thermal noise.

Noise was sampled using an aperture radius of 10 pixels hence giving a sample size of 314 pixels. This is very much in accord with BS 6001 for a sample size for a large batch.

The results are tabulated below. SD Mask give a very curious result in that it appears to smooth the noise to an extreme extent. I intend to do this using real light frames just as soon as the sky clears. Average is doing roughly what it should, ie; the noise is reducing as the square root of the number of frames combined.

Changing the settings for SD Mask had a limited effect. Increasing the sigma factor reduced the noise count slightly but given we are dealing with random noise the difference is likely to be lost ‘in the noise’. In other words, the values found are unlikely to be repeatable within 0.5 standard deviations.

Changing the number of passes made a significant difference to the time taken to process. The times given here are those taken to combine 12 ST10 images of 6.3MB each.

Further work needs to be done to properly characterise SD Mask but nothing I have seen so far has done anything to put me off using it all the time.

Any comments/criticism would be welcome.

Dennis

[MartinB]

Interesting that SD mask reduces noise compared with average, and by so much. This suggests there are some significant outliers. I wonder if this would be the case with real life images? Will be interested to see the further results Dennis.

I'm actually mulling over trying out CCDstack, helped by Adam Blocks video tutorials. At the moment though there seems little point since my weather just doesn't allow me to do any imaging and going over old images seems so sad!!

[roundycat]

Martin, I absolutely agree. If I feel inclined to go over any more old pictures I think I will delete the lot.

I still have a trial CCD Stack zipped up and I have no excuse not to open it up and give it a go. Well, one excuse really; most of my recent stuff has been calibrated, stacked and then saved and the raw files dumped so I have nothing to deal with.

The SD Mask result above is very odd, I'm hoping someone can come up with some suggestions as to what is happening. I can only think it is down to my using a small synthetic image but at the end of the day the algorithm does the same regardless of the size of the picture. It is combining pixels across the (16) frames one pixel at a time. It is only the measurement that is taking place on a sample of 314 pixels in each frame.

Thinks: I'll try it with a much smaller measurement aperture.

Dennis

[roundycat]

I think I have it! the gaussian noise I used goes from 0 to 126 adu. It is the zero values that may be skewing the result.

Dennis

[roundycat]

[removed word] it! Wrong again! Tried two different levels of noise and SD Mask is killing it all.

Dennis

[MartinB]

Nothing to suggest Dennis. You are a long way ahead of me

[roundycat]

managed to get some new frames to work on tonight. A quick grab of M46 has given me 21 fresh un-calibrated lights to play with. Watch this space!!

Dennis

[steppenwolf]

This is a fascinating article and very useful. I tend to discard all images with satellite trails and so on and my sensor is pretty 'quiet' so Sum works well for me but having seen the empirical test with Sigma Clip (which I have used on occasion but it takes soooo long on my PC!) I am tempted to keep more frames and try this combine mode instead.

What ancillary setting did you use for the Sigma Clip>

Thanks for sharing.

[roundycat]

Steppenwolf, the Sigma Clip values were 0.5, Linear and 50%.

I have just been playing around with the M46 frames I took last night. The results were quite different from the synthetic frames I was playing with (shown above in the white box) so I am led to believe that that type of test is invalid.

A set of 12 'proper' frames combined using Average gave me a noise, measured as Standard Deviation in one specific place, of 13.9. The equivalent SD Mask value was 10.3. This was using my standard settings of 3 passes and Sigma value 0.5.

Given that one pass with SD Mask made virtually no difference (0.2) and 2, 3 and 4 passes all gave the same value I will be using one pass in future. The reason for this is that increasing the number of passes significantly knocks down the highest pixel values. That must be one of the contributory reasons why the SD Mask values above are so small after combining a lot of frames. It has the associated benefit of being much quicker. I looked carefully at my M46 frames after different numbers of passes and could see no real difference.

Average is widely touted as the 'quietest' method of combining but with SD Mask knocking out the outliers and averaging the rest it seems self evident that it could end up giving a quieter result. As the algorithm looks at individual pixels across the frame set it will only knock out the bad ones if they are a single appearance. Hence there should be no real effect on any of the signal as that should be constant across the frame set. Pixels are only being dealt with on an individual basis anyway.

I shall use SD Mask settings of one pass, 0.5 sigma, 50% and ignore black pixels from now on and enjoy the benefit of doing my frame combining in half the time.

Dennis

[CELESCOPE]

Cant offer much here , Dennis seems to have gone deep into this , my findings from old using maxim , that SD mask seemed my best option , thus eleminating all hot pixels from my frames , after reading what dennis had done i tried all the various methods , of Maxim , and still found the results far less pleasing , finally finishing with SD MASK .

Maybe this is better for Mono then , cant say , i just know and feel that for me anyway SD Mask is what i like to use ,

However i would like on our next visit to Suffolk local meeting,maybe for Dennis to give some examples, always willing to learn .

Rog

[roundycat]

Rog,

how about I put together a short talk on what I have been doing with this combining lark and outline how I intend to go about calibration and combination in the future. Do you think any of the SGL EA members would be interested?

Dennis

[centroid]

I certainly would Dennis, and I'm sure it would also be of interest all.

Dave

Edited February 23, 2009 by centroid

[CELESCOPE]

Yes of course Dennis,thats what we are all about, we do tend to get into our own ways of processing wether that is good or not, so any new info would be very useful,

Please carry on ,

Cheers

Rog