Reduction of Noise in Astronomical CCD Images
Noise is caused by quantum (random) variations in the electric charge of each
CCD pixel (photosite) leading to grainy or speckled images. Here we look at ways to minimise
the effect.
Cooling
The number of unwanted electrons generated inside a CCD can be minimised
by keeping it cool. Purpose-built astronomical CCD's are usually mounted
on a cooling device, and noise from these CCD's can often be
ignored. Un-cooled cameras, ordinary digital cameras and webcams usually
generate quite a lot of noise. Some people put their webcams in the freezer
prior to using them, but beware of condensation!
The message is - if you can afford it buy a cooled camera!
Long Exposure
Long exposures will give time for the light from the target celestial
object to build up a strong charge relative to the noise. How
long is 'long' will very much depend on the magnitude of the subject and the
type of camera.
However, long exposures have their own problems:
- The brighter stars may overexpose (saturate) their pixels causing
distortion.
- Imperfections in telescope tracking can cause fuzziness or streaks.
For ordinary digital cameras and modified (long exposure) webcams a lot
of noise can be generated and there is a limit to how long the exposure can
continue before the whole image is 'fuzzed out'. You will need to experiment
to find out the optimum exposure. Unmodified webcams should be run with minimum 'gain' as the amplifiers
used to magnify their output introduce even more noise. I normally run my
Toucam at 5 or 10 frames per second and set the gain to as low as possible.
With a cooled camera you should make the exposure as long as possible up
to the point where poor tracking begins to fuzz the image or bright objects
become overexposed.
Stacking
Noise can be reduced by stacking (averaging) multiple images. The amount
of noise is reduced by a factor that is the square root of the number of
images averaged.
- Average 4 images - noise halved
- Average 100 images - noise reduced to 1/10th
This is where webcams benefit. Hundreds of video frames can be stacked to
reduce noise, (even though the webcam is quite noisy in each individual
frame).
In practice with my MX5C and DSI cameras I use exposures of up to 120seconds,
but this means I have to spend quite a bit of time to ensure the tracking is
good. I would try to collect 30-40 good images then stack them using
Registax or Maxim DL.
With the webcam I would generally take videos of 2minutes at 5 or 10 frames per
second and using the minimum gain possible. The video is stacked using Registax.
Filtering
Having obtained the best possible images from optimum exposure,
calibration and
stacking multiple images, the result can often be improved by filters
provided in your image processing software. Different software products use
different names for filters, but the two basic ones are:
|
|
Works by reducing the high-frequency (fine detail)
components of the image. In effect this smoothes the image making noise
less visible but at the expense of loss of detail or sharpness. This is
often fine for a nebula or galaxy images but not so good if you are
trying to get crisp clear lunar landscapes. |
|
|
Works by replacing the value of each pixel by the median
of the values of the pixels around it. This can be quite effective at
removing noise while maintaining a reasonable level of sharpness. |
Image processing software often has special noise reduction filters. PSP's
'edge preserving smooth' is one of my
favourites. It is quite good at smoothing a nebula while
maintaining the sharpness of stars. Astroart provides
the ability to view and edit an FFT transformation of the image and this can
be useful in detecting and correcting noise. The software package Neat Image
has advanced noise reduction capability and I use it quite often.
It is worth trying out several different noise reduction filters on an
image to see which works best.
Noise Reduction Strategy
My strategy is:
Toucam: Use the lowest gain possible. Capture at least 600 frames. Stack
only frames that are reasonably good. It is really worth the effort to
examine each frame of the video and select/deselect it rather than relying
on the software to select the best 'quality' frames. The automatic 'quality' functions
rely on a measure of image contrast but can include frames that have a lot
of shape-distortion in them.
Do not over-do the wavelet filtering (or other sharpening filter)
as this can accentuate the noise, (but do try the built-in noise reduction
function). Use PSP's 'edge preserving smooth' or Neat Image on the resulting
image.
MX5-C/DSI: Use exposure sufficient to capture a reasonable image without over
exposing individual stars or causing tracking problems. In practice this
means up to about 120secs per image with my equipment. Capture at least 10 images,
but 30 or 40 is better. Calibrate
with dark frames, colour convert (for MX5-C) then stack (average) the images. Experiment with
median filter, 'edge preserving smooth' and Neat Image to see which gives
the best result. It may be worth applying a median filter or FFT-Lowpass to
each individual exposed image before stacking.
There is often a trade-off between image sharpness and noise. Using an
UnSharp or Deconvolution filter to improve detail will tend to accentuate
noise. Several repetitions of moderate sharpening/de-noising may work
better than a single pass through each filter.
    |
