A very warty M42

[MarkS]

I took this on Saturday morning as an experiment.  It is very warty because the moon was creating a lot of background light.  The image also suffers from the classic "banding" problem exhibited by DSLRs.  It also suffers from that annoying glare around the bright stars.  I took some shorter subs to fill in the saturated parts but I haven't used them in this quick stack.  I think the banding problem can probably be dealt with because the fixed pattern appears to be unchanging from sub to sub.  So if I can somehow extract this fixed pattern from the subs I can then remove it from each sub during preprocessing.

As usual, H-alpha modded Canon 350D on Tak Epsilon 180ED
51 x 5 mins at ISO 800



Larger version here:
http://www.markshelley.co.uk/Astronomy/2011/hhm42_21102011.jpg

Mark

[mickw]

The large version really shows up the banding and I must admit I had never really paid much attention to it.

I was thinking that the bayer matrix was the culprit as the bands are more pronounced in particular colours (brownish), but that should also cause horizontal bands.
Could it be the way the camera processes the data (Digic 2 on the 350), does this still occur on newer cameras ?

[MarkS]

The banding is only obvious in the areas with very low signal levels (the brown dust in this particular image) where I've applied extreme stretch.  It's nothing to do with the Bayer but to do with the way the pixels are read from the CCD.  I've seen images that show it certainly affects the 300D, 350D, 400D, 450D, 500D, 550D and 1000D models.   It affects higher end models as well.  Just Google for something like "Canon 450D banding".

It's also known as fixed pattern noise.  The thing is, though, it is not fixed.  It is probably fixed on one sequence of shooting but if I compare the banding with earlier sets of images, I can see that the position of the bands definitely changes - at least for the 350D.

Mark

[RobertM]

It certainly a big annoyance that banding.  Did you find you whether it was in lights/darks and flats or peculiar to one in particular ?

Nice experimental image btw.

[MarkS]

Did you find you whether it was in lights/darks and flats or peculiar to one in particular ?

It's in the lights - I didn't shoot any darks or flats.  If I subtract any 2 lights at random, the banding disappears - i.e. the banding was constant over that whole shooting session (many hours).  

I repeated it last night, this time using a CLS filter, but there was no detectable banding.

I've uploaded a couple of subs to show the issue.  To highlight the banding I've subtracted a Gaussian blurred version of the sub from the original.

http://www.markshelley.co.uk/webdisk/m42withbanding.jpg
http://www.markshelley.co.uk/webdisk/m42nobanding.jpg

I've processed both subs in exactly the same way.

The one with banding was from Friday night with no CLS filter.
The one with no banding was from Sunday night with a CLS filter.

Mark

[JohnP]

Mark I remember when I used to use my modded toucam I always used to get noise bands (perhaps from interference or something) - There was a prog I used (might have even been PixInsight) where you could view image in Frequency domain - you could then use then pen tool with black ink & basically 'black' out the erroneous frequencies that was causing the banding - then when you went back to image view the noise had magically disappeared. Maybe Robert knows if this is a function of PI?

Cheers,  John

[RobertM]

That's a very intriging solution to the problem if possible.  I certainly don't know of any functionality in PI for the 'lines' problem apart from the 'Canon banding reduction' tool which only seems to have a very limited effect.

[MarkS]

I tried the idea of "blacking out" unwanted areas in the frequency domain but this, in turn, causes streaks to come out of the bright stars.  They would need masking off which isn't at all straightforward.

So I tried a different approach.  I've already determined that the streakiness is (luckily) constant across all subs taken that night. So I  performed a straight sum of the 50 raws and then subtracted a blurred version (5 pixel Guassian) of the same sum.  After converting to RGB, the result is below (this is a crop):



The "seagulls" are stars which have been very dithered during acquisition.

The streaks have now become pretty obvious.  It is also apparent that they are constant from top to bottom of the image i.e. they do not vary according to the local intensity of the image.  I don't know why this should be the case - it is probably something to do with how the data are read from the CCD pixels.

However, it then becomes possible to isolate the streaks and subtract them.  The result in the same crop is below:



So now we've succeeded in isolating the streaks, they can be subtracted from each sub before any processing begins.

Below is a before and after image:


Before:



After:



There is still some residual streakiness but these are streaks of wider amplitude and are not so annoying.  Changing the width of the Gaussian used above (from 5 pixels to, say, 10 pixels) might help further reduce the wider frequency streaking.

By the way the streaks that I isolated had an amplitude of +/- 10 ADUs which is quite large but hardly visible above the read noise and background noise until they were all stacked together.

Mark

[JohnP]

Mark,

I just remembered that Mr Noel Carboni Astro Tools for PhotoShop (Yes he of Star Spikes fame) has a tool for horizontal & Vertical band removal - I put your image in Pshop ran the vertical band removal tool & hey presto 2 secs later result is as shown below (full size crop) - If you want to see full image it is in link below...

So looks like Mr Carboni has got his uses...

Cheers,  John.



http://jpastronomy.co.uk/graphics/web/Mark-streaks%20removed.jpg

[mickw]

So looks like Mr Carboni has got his uses...

Remove "Diffraction Stripes" ?   

[MarkS]

John,

That is a very impressive stripe removal!

I wonder how it works?

Mark

[JohnP]

Well there is about 29 steps in the action - If I new a way to list them without having write them all down I would show you....

[JohnP]

Mark - have sent you an email - John

[MarkS]

I have decided my IRIS script works much better on the final image rather than applied to the subs.  I've also tried it out on my earliest ever horsehead image:

Before:


After:


Mark

[Carole]

Have read this with interest Mark.

I can't say I have noticed any vertical lines like this with my Canon DSLR, except for those horrid streaks I sometimes get after processing, but they are not straight like these and I am sure are something to do with DSS stacking when I do multiple sessions sometimes.  Do you think it is limited to certain models of Canon?

Carole