Dark Frames and Bias Frames Demystified – Sky & Telescope


One of the keys to facilitating image post processing is to record better data in the first place. I’ve already talked a lot about fundamental techniques to help you capture the best data possible and understand the limits of your equipment or the weather. Once you’ve collected your images though, you need to calibrate them to obtain the best results.

Calibration Progression
There is nothing wrong with your camera. Proper calibration is always needed for low light images.
Richard S. Wright Jr.

Image calibration is the first step of post processing, and when it’s done right it makes subsequent adjustments easier. Calibration helps remove artifacts that come with the image-acquisition process, so that your post processing deals with the actual good data you have worked so hard to acquire. Image calibration is also called data reduction, because it reduces all that you have collected to just the “data” part.

Many imagers skip calibration completely, and some do it improperly. Skipping a step can cost you time and effort later, and doing it improperly can make your initial starting point even worse than not doing it at all. Once images are clean, they require only minimal processing and produce stunning, informative, and honest images.

Pattern noise in an image
This is simulated, but I’ve seen worse. Faint signal stretched hard will bring out your sensor’s dark fixed-pattern noise. Proper calibration can help a great deal with this.
Richard S. Wright Jr.

To remove the artifacts of the camera and optical system from our data, we use three different kinds of master calibration frames. You’ve probably heard of them: bias, darks, and flats. Flats are important enough to get a blog all their own, so this month I’m going to focus on biases and darks.

Bias Frames

A bias frame is an image taken with no light falling on the image sensor, using the shortest exposure time you can manage with your camera. Either close the shutter or cap your telescope. Bias frames should be recorded at the same temperature as your light frames (the actual exposure of your target), and using all the same camera gain or ISO settings.

If you take your biases during the day, be careful that there are no light leaks getting to your sensor. Filter wheels and focusers often leak ambient light into your camera, which will ruin your bias frames. When I need to record bias frames during the day, I wrap much of the imaging train up with aluminum foil to keep this from happening.

Master Bias Frame
Bias frames capture dark fixed-pattern noise, shown here, from variations in manufacturing that affects all image sensors to some degree.
Richard S. Wright Jr.

Every image sensor, be it a CCD or CMOS, has what is known as dark fixed-pattern noise, a pattern that is the result of the manufacturing process. Every image you take records this faint pattern, no matter how long the exposure was or how much signal falls on your image sensor. The pattern then shows up in your images when you start to stretch (or brighten) the areas of your picture that collected little light.

To remove dark fixed-pattern noise, subtract a bias calibration image from your light image. In order for this step to work well, a master bias frame is created by stacking many individual bias frames, which removes the read noise. You can subtract the master bias frame from any image you take with that camera, with whatever length exposure, as long as the other camera settings (temperature, gain, offset, etc.) are the same.

Dark Frames

A dark frame is like a bias frame in that it’s an image taken with no light falling on the image sensor, but dark frames need to be the same length as your light frames. In other words, if you take several 3-minute exposures on your target, you’ll want to calibrate them using a 3-minute master dark frame, which you’ll subtract from the image. This calibration step removes two things: First, your master dark contains the same dark fixed-pattern noise that your master bias frame does. It also collects dark current, and more pattern noise called DSNU (Dark Signal Non Uniformity). Individual dark frames also contain their associated shot noise with that comes along the dark current.

If you use a master dark frame, you don’t need a master bias frame — you really don’t want to subtract the dark fixed-pattern twice!

Uncooled vs. Cooled
The left image was recorded without cooling and suffers from excessive noise from the resulting dark current. Richard S. Wright Jr.

The dark current comes from thermal activity (that is, heat) in the image sensor, and it creates a growing offset to all our pixel values that increases with both time and higher temperatures. If the effect were uniform we might not mind so much, but the offset is spread randomly among the pixels (the DSNU). The dark current also feeds “hot pixels” — pixels that appear much brighter than their neighbors. A good master dark can do a lot to remove that salty appearance from your raw frames. Cooling the sensor also greatly reduces the thermal current that pollutes images.

We can’t simply subtract the shot noise associated with dark current from the dark frame; instead, we have to stack dark frames to minimize the noise. That way, this random noise doesn’t pollute all the light frames that we’re calibrating. The dark current’s shot noise is also in our light frames, but we can only remove this noise by stacking lot of light frames. When we subtract a dark frame, we remove hot pixel offsets and the dark current offset, but we can’t subtract the dark current’s shot noise — stacking is the only way to remove shot noise of any kind.

Bad Pixels
Hot pixels can detract from a monochrome or color image. There are many techniques for removing them, but dark frames are a good first defense.
Richard S. Wright Jr.

The Future

So why talk about bias frames if all you really need is a dark frame? Because technological developments may eventually make dark frames obsolete. There are many newer image sensors with extremely low dark current when cooled sufficiently. I really hope this trend continues. Once cooled they may gain a single electron or less per pixel over long periods of time — even 20 minutes in one sensor I’ve tested.

If the camera sensor has no appreciative dark current when cooled, you can apply bias frames to your data and skip doing darks all together. You may still get some hot pixels here and there with these cameras, but those are easily removed with a pixel map in post processing or by dithering your exposures and stacking with a rejection algorithm.

Some CMOS sensors also actively drain off dark current as it accumulates. You can watch this happen by taking longer and longer dark frames and observing that no additional background signal accumulates, even at warm temperatures! Again, in these cases, a good clean bias frame is all you’ll really need, plus stacking plenty of individual exposures.

Stay tuned: Next time I’m going to talk about the alchemy of flat-frame calibration and why often people have such a hard time getting them to work properly for them.





Source link