Dark frames?

Hey Astrodudes,

So, every time I try to take a "dark frame", I take it at the same settings as I do my light frame. To absolutely no avail, it does not decrease the noise whatsoever--it only increases it.

None of the tutorials online really make much sense. Could someone explain things a bit better to me? What am I doing wrong?

The dark frame method is to remove hot pixels not noise.

If you have lightroom or Photoshop it'll automatically remove them for you.

Again hot pixels not noise. For noise try native ISO with magic lantern, it makes a huge difference.

Annnnddd now I feel like an idiot.

So what's the point of having like 40 light frames and 10 dark frames? I'm confused about the theory of it.

Am I better off shooting light frames at a higher ISO and shorter shutter speed, and taking more photos, and then taking darks, as opposed to taking less light frames with a lower ISO and slower shutter speed and taking less photos?


And how does one stack them on a Mac? Do they just use photoshop or something similar to DSS?

Errm no, dark frames are to reduce noise, not sure what you are doing wrong though,
http://en.m.wikipedia.​org/wiki/Dark-frame_subtraction

There are two types of noise - random and systematic.

Let's imagine you're shooting with a sensor that has a range of two bits - each pixel is on or off, 1 or 0. Where your image contains a real light source the pixel will be on (1), where it's dark it'll be off (0). Random noise will mean that any pixel has a chance of lighting up even though there's no light hitting it. Let's say the chance is 10%. So your single image has lots of stars - 1s - and lots of dark sky - 0s - but it will also have 10% of that dark sky as noise - also 1s. You have a signal:noise of 1:1

Now let's take 10 such photos and stack them - add all the individual signals together. All of the stars will be in the same place - so each star will have a value of 10. However, because the noise is random the chances that a noisy pixel in one image coincides with a noisy pixel in one of the other 9 images is about 50:50 - so you'll have a lot of pixels with a value of 1, several with a value of 2, a few with a value of three, etc. You can see that your signal:noise has improved significantly, close to 5:1 Now stack 100 images. Your stars will all have a value of 100 - but your noise pixels will only have increased a little bit. Indeed, the signal:noise ratio will be around 25:1 (although I fear I may be out in my maths - but the general idea is correct).

So, that's how to reduce random noise. What about systematic noise - hot pixels?

Hot pixels are always on - on our imaginary sensor they always give a signal of 1. Canon's Long-Exposure Noise Reduction takes a dark frame by making an exposure using the same settings without opening the shutter - then subtracting that image from your light-frame. Which is OK - except that the dark-frame image will also have some random noise - which will also get subtracted from your light-frame. This is not drastic (you'll have some extra dark pixels) but it can be avoided.

If you take several dark frames then stacking them will result in large signal for the hot pixels (always in the same place) and much smaller signals for the random noise (sacettered in ransom locations). So the stacking software will only remove those bright pixels.

Good question.

First reply is not accurate. That is referring to Bias Frame (Also called Offset frame) and not Dark frame.

Frank's reply is correct as usual, if you can understand that.

Before going any further, let me ask you: What have you understand so far? Express it and you will get more help if you are not right.

Alright. For example, I drove about 20 minutes out of town to a darkish area to get away from lights so that I could see the stars. I took 20 light frames at the same settings, put my lens cap on, and took 20 more pictures at the same settings as the light frames were taken. I attempted to stack them, but DSS came out with whatever excuse it could--"out of memory" most of the time (eye roll) to stop the process 3/4ths of the way through. So I gave up. But do I have the right concept?

I just don't get how people get these results--but when I do it, I don't really get the nebulas and stuff. There's something I'm missing. Is it photoshopped in?

http://www.weasner.com …may08/OO_08_May​_31_93.jpg

quadwing wrote in post #16492955
Alright. For example, I drove about 20 minutes out of town to a darkish area to get away from lights so that I could see the stars. I took 20 light frames at the same settings, put my lens cap on, and took 20 more pictures at the same settings as the light frames were taken. I attempted to stack them, but DSS came out with whatever excuse it could--"out of memory" most of the time (eye roll) to stop the process 3/4ths of the way through. So I gave up. But do I have the right concept?

I just don't get how people get these results--but when I do it, I don't really get the nebulas and stuff. There's something I'm missing. Is it photoshopped in?

http://www.weasner.com …may08/OO_08_May​_31_93.jpg

As already stated above, you'll need to subtract the dark frames from the light frames stack to reduce noise.

In order to resolve the dust and nebula with the Milky Way you'll need to have individual light frames that contain enough data to work well in the stack. This means stretching your exposure times as long as possible(short of visible star movement) and do so at an ISO setting that you're comfortable with. ISO 800 is a good starting point in moderate outside temps, in cold conditions 1600 should work well. If the skies at the location you mentioned are fairly dark then you should be in good shape. I'm assuming you're shooting from a tripod and not a tracking mount so only by testing exposure times at the focal length you're shooting can you find the optimal length of time to get the best light frames. You can't stack extremely short light frame exposures and expect the final stacked image to be stunning. The final stacked image is only as good as the individual frames used. Stacking does not create data it raises the signal as Frank explained but if the data in each frame is lacking then the stack will reflect the deficiency. So the key is to gather the most light possible within your restrictions of focal length and sky conditions. The easiest way to do this is with a tracking device under very dark skies but it can be done from a standard tripod with some practice.

I think there's an option called 2xdrizzle or 3xdrizzle that whenever was ticked in dss always gave me the out of memory error. Haven't got dss to hand so maybe someone else can confirm. Also as I recall you only need about 20% dark frames to light frames.

Coming to this late.. I notice one or two points not covered.

Dark frames.. to reduce noise you need to do two things, as others have said, you need to average multiple dark frames so the random noise doesn't noticeably add.. I would regard 8 dark frames as an absolute minimum.

Sensor Temperature. The Dark frames should be taken with the sensor as close as possible in temperature to the light frames... don't take the camera indoors to take darks.

Problem: Using the sensor warms it up.. so you can never get exactly the same temperature and it all starts getting complicated and simply not working (I've even had problems with a set-point cooled camera)

Solution: Don't just subtract the dark frame. Use a correlated subtraction: A process where the dark is multiplied by a constant before subtraction so there is zero correlation between the final result and your dark frame. You still need your darks taken close in temperature to your lights, but a correlated subtraction makes a noticeable difference in my experience (both on DSLR and CCD)

Software?.. Personally I don't use DSS.. it may have a correlated subtraction option, I just don't know. I use IRIS and in that there is an 'OPT' command which does precisely this, it works amazingly well.

Alternative?

Hot pixel removal. Generate a 'Hot Pixel Map' from your dark frame, use that to 'fix' the light frames.. I've had surprising levels of success by just doing that and not subtracting the dark frame at all, I'm 99% sure DSS will do it.

Derek

rfdesigner wrote in post #16503073
Coming to this late.. I notice one or two points not covered.

Dark frames.. to reduce noise you need to do two things, as others have said, you need to average multiple dark frames so the random noise doesn't noticeably add.. I would regard 8 dark frames as an absolute minimum.

Sensor Temperature. The Dark frames should be taken with the sensor as close as possible in temperature to the light frames... don't take the camera indoors to take darks.

Problem: Using the sensor warms it up.. so you can never get exactly the same temperature and it all starts getting complicated and simply not working (I've even had problems with a set-point cooled camera)

Solution: Don't just subtract the dark frame. Use a correlated subtraction: A process where the dark is multiplied by a constant before subtraction so there is zero correlation between the final result and your dark frame. You still need your darks taken close in temperature to your lights, but a correlated subtraction makes a noticeable difference in my experience (both on DSLR and CCD)

Software?.. Personally I don't use DSS.. it may have a correlated subtraction option, I just don't know. I use IRIS and in that there is an 'OPT' command which does precisely this, it works amazingly well.

Alternative?

Hot pixel removal. Generate a 'Hot Pixel Map' from your dark frame, use that to 'fix' the light frames.. I've had surprising levels of success by just doing that and not subtracting the dark frame at all, I'm 99% sure DSS will do it.

Derek

How does someone go about using correlated subtraction?

quadwing wrote in post #16504031
How does someone go about using correlated subtraction?

Sorry, I've been very busy and not logged in for some time.. so I just picked up your question.

I use IRIS. It has a functions called 'OPT' or similar which do the correlated subtraction. If you go here :http://www.astrosurf.c​om …ris/tutorial2/d​oc9_us.htm there are some instructions on doing it.

IRIS will load CR2 files.. although it works in singed 16 bit/colour .FIT images..

Download for free here:

http://www.astrosurf.c​om/buil/us/iris/iris.h​tm

If all you do is use it for dark subtraction it's worth it.. I use it for the full flow but it's hard to get to grips with, and I don't direct people to use it for what I do as I get a lot of 'too hard to use' type responses. People who use it to do pre-processing swear by it though.

The real way to drive IRIS is with the consol and command lines, which means having the commands list printed out.. that's how I drive it and I find it very powerful.

Derek

quadwing - I had the same problem with DSS and it turned out that I was not using the exact same exposure time and ISO for the dark frames as the light frames. The second time I tried I inadvertently included a bias frame in with the dark frames. Same error messages. Once I found MY error everything worked.

Thanks to everyone else for all the good information.

Out of memory is a common problem with DSS. It means what it says. You will need more RAM for your computer to handle this massive amount of data processing.

You are doing the right thing. Take 20 or thirty light frames, take ten or twenty dark frames, ten or twenty flats, the same in offset/bias frames (fastest shutter speed your camera can take) and throw them all in DSS and let it work it's magic. Once it's aligned and stacked everything, then all you have to do is play with the sliders to bring out the data and export as a TIFF for tweaking in photoshop.

Easy peasy.

Of course, if your RAM is too small, you're screwed.....

I wrote this tutoriala long time ago, it's a little dated but you should get the idea.

Baz.