DSLR Noise reduction

[Pompey Monkey]

I took this image of M51 a couple of weeks back: 14 x 10min lights at 1600 ISO, about 30 flats, 20 or so bias and 8 darks (in the fridge, post event, lol). Equipment as in my signature.

I stacked in DSS, using it to get the RGB channels equalised. The only processing taht I've done is to convert it to 16 bit (I chose the option that kept the detail in the core) and RC Xterminator to reduce the background graidient.

My question is: How do I reduce the noise in the image? I understand that I could always do with more data and that I did not set up the dithering between APT and EQMOD. That aside, what next?

Edit: My 4 MB jpeg has been compressed to about 20 bytes by the website How do I upload an image so that you can see the noise?

Edit 2: attached as zipped file.

Autosave006_merged_11.zip

[Ruud]

Post it as a zip, or try dropbox and post a link to the file.

I use Topaz DeNoise myself, but that is not an astronomy specific tool.

[Pompey Monkey]

Post it as a zip, or try dropbox and post a link to the file.

I use Topaz DeNoise myself, but that is not an astronomy specific tool.

Thanks Ruud for the tip about attaching a zipped file. - I will look into the Tpaz DeNoise,

[Pompey Monkey]

Thanks Ruud for the tip about attaching a zipped file. - I will look into the Tpaz DeNoise,

Buy Now for $79.99

No thanks! lol.

I tried it too. It seems to have an effect similar to using a powered rotary wire brush on a Louis 14th plaster moulding!...

...Maybe I just prefer the more raw look, but thanks again for the heads-up!

[Traveller]

10 mins seem a long single exposure, esp at 1600 ISO. I tend to shot 5 min at 800 ISO. More data will help. Have you thought about getting a cooler box? That will reduce noise and allow fr longer exposures.

Bo 

[Pompey Monkey]

10 mins seem a long single exposure, esp at 1600 ISO. I tend to shot 5 min at 800 ISO. More data will help. Have you thought about getting a cooler box? That will reduce noise and allow fr longer exposures.

Bo 

My histogram is sitting at about 25% - some would say the sweet spot. I am using an Astronomik CLS filter, which reduces the transmitted light a fair amount.

Yes. I am experimenting between 800 and 1600 ISO - there does not seem to be much difference in the noise, at least subjectively. 

I have just looked at the cooled DSLR cameras - this is not a cheap option. I can feel a DIY project forming in my head...

[swag72]

For me, I tried dithering and wrapping ice packs around the DSLR, in the end my solution was to move to a cooled CCD.

[wxsatuser]

My histogram is sitting at about 25% - some would say the sweet spot. I am using an Astronomik CLS filter, which reduces the transmitted light a fair amount.

Yes. I am experimenting between 800 and 1600 ISO - there does not seem to be much difference in the noise, at least subjectively. 

I have just looked at the cooled DSLR cameras - this is not a cheap option. I can feel a DIY project forming in my head...

It looks like the " Walking Noise " talked about on CN.

Dithering is supposed to put paid to it.

Never had it myself apart from once when I did 4hours on M101 using ISO400.

I have'nt had it since using mainly ISO 1600 and I don't dither.

For noise I use ACDNR in PixInsight and Deep space noise reduction in Noel Carboni's stuff.

I also try to get plenty of good data.

[Pompey Monkey]

It looks like the " Walking Noise " talked about on CN.

Dithering is supposed to put paid to it.

Never had it myself apart from once when I did 4hours on M101 using ISO400.

I have'nt had it since using mainly ISO 1600 and I don't dither.

For noise I use ACDNR in PixInsight and Deep space noise reduction in Noel Carboni's stuff.

I also try to get plenty of good data.

Yes, It does look like that "walking/correlated noise" discussed on CN (and thanks for the search term - I'd never have found the thread by accident!). The fact that aggressive dithering seems to cure it ties in nicely with the paper that Craig Stark wrote about the non-linearity of the noise in Canon DSLR dark frames. There is certainly some manipulation of the data in the camera before the "RAW" file is created. 

I meant to dither - It just turned out that I had not correctly set up APT with EQMOD to do it (first time user!).

I've got the Carboni plugin - I'll try the "deep space noise reduction" when I get home tonight - I'm still learning how to use these tools too.

I've only been doing this for about three months and I want to see how to get it right with the gear that I've got now before spluging out on an expensive CCD. Although I'm sure that I will one day! lol

[wxsatuser]

May be I should dither but I use an intervalometer so it's not practical.

I looked for a long time on what type of noise this was and only found it recently.

[Pompey Monkey]

I've been thinking about this, as you do:

It occurred to me that If I am aggressively dithering my lights and Canon is doing some processing to the so-called "RAW" files, then there is no point in using dark frames. I just need to make sure that I have the bias and flats.

Indeed, this has been discussed before: http://stargazerslounge.com/topic/223975-dslr-imaging-do-you-really-need-dark-frames/

I will experiment next time I get a chance by stacking with and without the darks to see what happens. I'd do it today, but I don't have any dithered lights to play with!

[wxsatuser]

I understand that after 2008 Canon do on sensor dark current suppression.

Now I don't know if all Canons have it but I did see someone else, somewhere, post about the 60Da

and they did a comparison,  to me darks aren't needed.

All the hot/dead/stuck pixels are taken care of by doing a sensor clean before an imaging run.

This updates the pixel map and the RAW converter should sort these out.

I believe if we can get at least 30 frames of 300secs or more the image is going to be pretty good noise wise.

[Thalestris24]

I understand that after 2008 Canon do on sensor dark current suppression.

Now I don't know if all Canons have it but I did see someone else, somewhere, post about the 60Da

and they did a comparison,  to me darks aren't needed.

All the hot/dead/stuck pixels are taken care of by doing a sensor clean before an imaging run.

This updates the pixel map and the RAW converter should sort these out.

I believe if we can get at least 30 frames of 300secs or more the image is going to be pretty good noise wise.

Hi Mike

I believe that requires taking a dark frame immediately after the light so more than doubles the normal exposure time. Seems better to take sets of darks at a convenient time rather than wasting valuable imaging time?

Also, not all Canons have sensor cleaning e.g. the 1100d doesn't.

Louise

[wxsatuser]

It's not a dark frame, it's on sensor dark current suppression.

Re 1100d......as I said it's possible not all models have these features.

Would assume the more up market models have these features, the 60D seems to.

[RikM]

I found dithering was a huge help to suppressing noise and 30 subs seemed to be the magic number for me. This was using a 1000D on a 150P. I always used ISO1600 with that camera. After stretching to the same brightness there didn't seem any difference in noise between ISO1600 and ISO800, and I found it easier to process starting from ISO1600. (That's just me though). I took a big run of dark frames in the fridge at 5min and 7min which were my standard sub lengths and used them for about 18months. They seemed to help.

Noel's deep space noise reduction run as a new layer with the opacity dropped to 60-80% was a standard part of my DSLR workflow.

[wxsatuser]

There are various ways to overcome noise and I'm not suggesting this is a good way for everyone.

I have matched darks to lights and have scaled them but the result after stacking does'nt look much different than no darks.

Basically I don't bother with calibration frames and look to get as many subs as possible or the clouds allow.

[RikM]

Completely agree. There is no single 'right' way to do it. If you find a method that works for you, great

[tich]

Dithering? what's that? Sorry for the daft question.

[RikM]

Dithering is when you move the mount a few pixels between exposures. This means that got pixels are scattered all over the frames. When you stack them up using a sigma reject algorithm, the hot pixels get rejected. Many capture programs can talk to guiding programs and coordinate dithering between frames. I use Maxim with my CCD and APT with my DSLR.

[tich]

Dithering is when you move the mount a few pixels between exposures. This means that got pixels are scattered all over the frames. When you stack them up using a sigma reject algorithm, the hot pixels get rejected. Many capture programs can talk to guiding programs and coordinate dithering between frames. I use Maxim with my CCD and APT with my DSLR.

Cheers Rik, I use APT with my DSLR so I'll have to figure out how to use it. I'm guessing the sigma reject algorithm can be found in DSS.

[RikM]

Yes. It's called Kappa-Sigma clipping or something like that.

[lensman57]

I understand that after 2008 Canon do on sensor dark current suppression.

Now I don't know if all Canons have it but I did see someone else, somewhere, post about the 60Da

and they did a comparison,  to me darks aren't needed.

All the hot/dead/stuck pixels are taken care of by doing a sensor clean before an imaging run.

This updates the pixel map and the RAW converter should sort these out.

I believe if we can get at least 30 frames of 300secs or more the image is going to be pretty good noise wise.

I believe that 60Da and the 6D and perhaps other high end DSLRs do on chip dark subtraction as their long exposure output are quite clean for a DSLR. If this is correct then the problem it presents the imager is whether to subtract Bias or not and then there is the question of the Flats. Although I still use DSLRs for imaging I find that I am more and more comfortable with my CCDs.

A.G

[ollypenrice]

In terms of applying post processing noise reduction I think the golden rule is to apply it only where it's needed and never by too much. Noise is at its worst when signal is low, so the background sky is the obvious first candidate. Highly stretched faint nebulosity will also need NR. 

You can do this using brighness based masks (the PI method which can also be done in Ps) but I find this a real pain. In Ps I prefer to make a copy layer of the noisy image and give the bottom layer a heavy dose of NR. Then I activate the top layer and use the colour select tool to pick up the parts needing NR. I'll start with a big selection including background and faint nebulosity and then use a big eraser set at whatever opacity gives a good result on the brighter parts. Run that over the selected areas once. Then use the 'reduce selection' option (or deselect and reselect only the background sky without the nebulosity) and run the partial opacity eraser over it again. Maybe reduce the selection again and repeat. You can also use a smaller eraser to attack any locally bad noise.

I would always leave a little noise visible.

There are other tricks. Zoom into pixel scale and see if you have a noise speckling of a particular colour. If you do, use colour select to pick up a lot of the bad pixels at once and apply the Median filter to give them the same value as their neighbours.

Olly

BTW, if you can't get dithering to work you can always switch off the guide for a few seconds between subs then relaunch it for the next sub. Even if you don't do it every time it will be a big help. If I'm not mistaken a dither can be sub pixel, too. It doesn't have to be much to work.

[wxsatuser]

I believe that 60Da and the 6D and perhaps other high end DSLRs do on chip dark subtraction as their long exposure output are quite clean for a DSLR. If this is correct then the problem it presents the imager is whether to subtract Bias or not and then there is the question of the Flats. Although I still use DSLRs for imaging I find that I am more and more comfortable with my CCDs.

A.G

Thats fair enough, I can understand the attraction of the CCD.

According to Roger Clark darks and bias are not needed with one of these cameras.

I have tried with and without both but on a good set of lights I cannot see the difference, so I don't use any at all.

This is not to say there is a difference, just that it's not obvious to me.

[wxsatuser]

In terms of applying post processing noise reduction I think the golden rule is to apply it only where it's needed and never by too much. Noise is at its worst when signal is low, so the background sky is the obvious first candidate. Highly stretched faint nebulosity will also need NR. 

You can do this using brighness based masks (the PI method which can also be done in Ps) but I find this a real pain. In Ps I prefer to make a copy layer of the noisy image and give the bottom layer a heavy dose of NR. Then I activate the top layer and use the colour select tool to pick up the parts needing NR. I'll start with a big selection including background and faint nebulosity and then use a big eraser set at whatever opacity gives a good result on the brighter parts. Run that over the selected areas once. Then use the 'reduce selection' option (or deselect and reselect only the background sky without the nebulosity) and run the partial opacity eraser over it again. Maybe reduce the selection again and repeat. You can also use a smaller eraser to attack any locally bad noise.

I would always leave a little noise visible.

There are other tricks. Zoom into pixel scale and see if you have a noise speckling of a particular colour. If you do, use colour select to pick up a lot of the bad pixels at once and apply the Median filter to give them the same value as their neighbours.

Olly

BTW, if you can't get dithering to work you can always switch off the guide for a few seconds between subs then relaunch it for the next sub. Even if you don't do it every time it will be a big help. If I'm not mistaken a dither can be sub pixel, too. It doesn't have to be much to work.

Thanks Olly, will look into that dither option.

In my short experience with processing I try not to push anything to hard, if I can help it.

I have followed a few tutorials that included noise reduction but the results are not so good.

So now I just give my data some ACDNR in PI and some Noel Carboni deepsky stuff in PS.