(very) noisy Red layer, G and B layers just fine - how to handle this?

[rodolgo]

Hi everyone,

Using a non-defiltered 1100D Canon DSLR, most of my stacked and pre-processed images reveal a very noisy red layer (not surprisingly though).

Until I get my DSLR modified... do you have any hints on reducing the noise in the red layer, separately from the green and blue layers which are both OK?

Thanks!

Rodolphe

[pfile]

what do your red-channel histograms look like? if the back-of-camera histogram does not have the left edge of the hump 20-30% of the way across the histogram display, your red data is going to be stuck in the read noise of the sensor. you may have to increase your subexposure length to get your histogram there. just be careful that the longer exposures don't end up overexposing the G and B channels.

unfortunately after that, the only 'real' way to attack the noise is to go for longer total integration time, that is, more subs.

of course as you point out, modifying the camera will help a great deal. you can also try cooling the camera to cut down the thermal noise.

beyond that, i guess you'll have to do more aggressive noise reduction on the red channel.

[Nocturnal]

Hi,

If you want to do noise reduction on one channel you can simply split your image into the three channels, process whichever channels you want and then recombine.

[rodolgo]

Thanks for your good suggestions guys,

I checked the histogram, the red is not stuck to the left part. Increasing the sub exposures is clearly a way to go, I'll try that next time.

Splitting the channels is another option, but I'm not sure - when recombining, will the color balance be changed compared to before the split?

Thanks!

Rodolphe

[Nocturnal]

The color balance will be affected if you mess with the channels in such a way that they affect the color balance. I realize this is a recursive answer but it is that way on purpose. You already know the answer to your question: "it depends". To answer your next question: "yes you can fix your color balance later, noise levels do not affect color balance. Signal levels do."

[alvinjamur]

The histograms shown at the back of the camera comes from a converted raw file in an assigned
colorspace (sRGB) and IS NOT an accurate representation of the bin values in the raw file. At least
for my workflow and in classes I've taught in terrestrial photography, I ask people to shoot to the right ie., expose in a way that you are almost clipping values on your histograms in the camera - it will look terrible and overexposed when u see the image on camera but that can be fixed in image processing software (i use lightroom for terrestrial stuff). shooting to the right maximizes the signal/noise ratio in the raw file as CCDs are linear amplification devices. The same would hold for images taken by a canon/nikon etc., terrestrial camera.

....pellucid skies!!


aLV
www.alvinjamur.com

[pfile]

this is absolutely true but the 20-30% back-of-camera trick is a rule of thumb developed over time for canon DSLRs - an easy way to to ensure that your signal is above the read noise of the camera.

once you have done this, then total integration time determines the SNR of the final stacked image.

don't forget that a picture of the night sky usually contains some very bright objects that are not part of your target: foreground stars. if you expose your subs too long, you will overexpose the stars and lose star color in your final picture. of course you can recover them by shooting short subs and doing an HDR merge.

[rodolgo]

So, in summary, the consensus looks like this:

• subs should follow the 20-30% rule of thumb (histogram read from the back of the camera) - take enough to get a good SNR; bright stars, if any in the FOV, might be over-exposed
• if there are bright stars in the FOV, a series of shorter (faster) subs are taken to do the bright stars and not loose their color, in which case you would do an HDR composition to merge the 2 results

Correct?

[Nocturnal]

I would not worry about star color at this point. That just adds complication you do not need right now. Start by taking guided exposures with exposure lengths as long as you can manage. You'd be lucky to have exposures so long that your stars look bad because they're washed out. Yes, bright stars will max out your sensor. No big deal. The edges will still retain color. If you process correctly your stars will look nice.

- don't touch the white point in the histogram
- increase saturation in the bright areas of your image somewhere late in your processing steps

This will result in (in my opinion) pleasing soft stars with colored edges if they are really bright and nicely colored stars if they're dim.

Walk, then run.

[pfile]

i was merely trying to point out one reason why you might not want really long subexposures.

other reasons include:
 1 the longer the sub, the more chances for something to go wrong with tracking (wind, mount issues)
 2 the longer the sub, the higher the odds of an airplane or a satellite spoiling the sub
 3 the longer the sub, the more chance a cloud could ruin your sub.
 4 loss of dynamic range (okay, it's the same thing i mentioned before; the stars being the first thing that will saturate)

#2 isn't necessarily a killer since there are methods to exclude outlier pixels from your stack. but if you have 4 hours of integration time with 30 minute subs (8 subs), it might be hard to get rid of that airplane without throwing away the sub. on the other hand if you have 4 hours with 5 minute subs (48 subs) rejecting the right pixels is easier.

basically the back of histogram trick is a convenient rule of thumb for DSLRs. if you really want to know how long your subexposure should be, you need to characterize your camera (what is the read noise?) and characterize your sky brightness (how much skyfog do i have per unit time?). there is actually a script in pixinsight called CalculateSkyLimitedExposure that can help with this. but until you are running, as sander says, you might as well just use the rule of thumb.

[Nocturnal]

Yes, the optimum exposure length has lots of factors going in them and you list some excellent ones. In general though very few beginning astro photographers run into those issues as their exposure length is limited by differential flexure and other issues they do have control over. In other words it isn't worth worrying about 20 minute subs and how many will get ruined by airplanes when all you can do is 2 minute subs before the stars turn to ovals

For now the right advise for Rodolphe is to go for longer exposures and to remove the IR filter if possible. Once he's getting exposures that are long enough that he's catching planes and loosing star color, let's tackle that.

[rodolgo]

For now the right advise for Rodolphe is to go for longer exposures and to remove the IR filter if possible. Once he's getting exposures that are long enough that he's catching planes and loosing star color, let's tackle that.

That sounds a pragmatic and reasonable advise I just need some clear night sky to be able to test, as snow is not helping much right now!

Thanks Nocturnal and others helping me out in this thread, much appreciated.

Rodolphe

[Nocturnal]

Keep us posted Rodolphe!