How to Take a Good Dark and Flat Frame

John A McCubbin

Principles

• Dark frames represent chip noise and are temperature dependent
• Bias frames represent readout differences of each pixel
• Flat frames represent optical imperfections in the telescope and camera
• Image calibration involves the removal of these three unwanted effects from the image

Method

A raw CCD image represents the visual display of stored electrons in each pixel of the CCD camera chip. Those electrons come from three sources. The first source is the captured photons from the object being imaged that stimulate electron buildup in each pixel of the CCD chip. The second source is accumulation of electrons acquired from random electronic noise in each pixel element of the chip. This is related to the operating temperature of the camera. The third source of electron buildup comes from the fact that pixel elements have different "zero points" after readout.

Add to that that the actual photons striking the CCD chip that emanate from the object being imaged, are modified in their intensity by optical imperfections in the telescope and by dust on the lens, CCD camera, and any filters in between. In order to get the highest quality image out of the raw image we want to subtract all sources of noise from the raw image leaving only a representation of the photons from the object itself. Then you want to compensate for any variations in brightness caused by the optics of the telescope or dust in the optical path. That is why images are dark subtracted, bias subtracted, and flat framed.

Dark subtracting removes the electronic noise from the image that builds up during exposure from the electronics in the camera and the inherent noise in the CCD pixels themselves. A dark frame measures the dark current, or current that builds up in the pixel elements of the chip over time, even with the shutter closed (mostly from heat - that is why dark current is lower at lower temperatures). Each pixel on the chip will respond differently at any given temperature, and the dark frame represents this. If you look at a dark frame visually, it looks like a field of snow with bright and dim pixels side by side. Since it temperature dependent, you should cool you camera to the lowest level possible that doesn't utilize over 70% sustained power of the thermoelectric cooling circuits (over this induces more noise in my experience).

Dark subtracting is fairly simple, but not as straightforward as you might imagine. The general principle is to take an image with the camera shutter closed (or the end of the telescope occluded if your camera doesn't have a shutter) and mathematically subtract it from the raw image. The dark frame should be at the same resolution, time, and temperature as the image you subtract if from. The trick to a good dark frame is that it must be taken in almost the exact same setting as the image (temperature, power settings on the thermoelectric cooling circuits, etc.). Since dark current is at least slightly random, the best dark frame is one that is an average of several dark frames in the same setting. Here is where practicality wins out. It is impractical for me personally to take two or three 20-minute dark frames and average them. Even a 10 or 5 minute dark frame rarely gets averaged, but an average of two is better than a single frame. If you are doing photometry, then you almost have to average several dark frames. (See the article on image stacking for the technique on averaging images using CCD Soft.)

A bias frame is acquired by taking a zero length exposure (or as near it as possible) and downloading it. The bias frame is a representation of the noise that occurs during the download process, making the zero level of each pixel fluctuate slightly. It can be a small source of noise in the image. It isn't nearly the source of noise, however, that dark current is. Since this is the same from exposure to exposure, you simply have to insure that the binning settings match and subtract the bias frame from the raw image.

Flat framing is trickier, in my opinion. Flat framing is necessary since there are few telescopes that have a perfectly even illumination across the field when properly focused. There aren't many that don't have at least a speck or two of dust, and cameras also build up dust that casts shadows onto the chip. Dividing by the sensitivity of each pixel represented on the flat frame will eliminate the sources of optical distortion. One must be careful to not introduce noise in the image while flat framing. This is the reason I average (not add) at least four, and most of the time eight flat, and occasionally twelve frames to make a master flat prior to image calibration. This prevents the randomness in the flat frame appearing as "noise" or graniness in the final image. This is really important, so don't scrimp on this step (it is usually performed last, in my case, when I'm tired and cold - so I'm usually tempted to cut it short).

To get a proper flat frame you must take a short exposure of an evenly illuminated object. Some suggest using the evening sky, I work from an observatory and like to take them last. If you do them first, then you cannot remove the camera for any reason (including finding a difficult object), or the flat frames won't be any good. I use an old projector screen as the target. I illuminate it by shining a 75 watt bulb into the opposite wall of the observatory and letting the reflected light illuminate the projector screen. I'm careful that nothing casts a shadow onto the screen. For my particular camera if I'm in full resolution mode, I take a 6 to 8 second exposure with an automatic dark subtraction (don't forget to dark subtract these - that's a big source of noise). That yields a 10-12,000 count reading for each pixel. If I'm binned 2x2 I use 2-3 sec. If I'm at 3x3, I use 1 sec. I then take eight flat frames. I then average (actually blend) them using CCD Soft (see stacking techniques) to make a master flat frame. I use the master flat for image calibration. I find that this all but eliminates noise from the flat frame and can dramatically improve the image quality without introducing noise.

Calibrating your images using these techniques will result in dramatically improved image quality.

Back to Resources Page

I consider your heavens, the work of your fingers, the moon and the stars, which you have set in place ... Psalms 8:3