| article: Thoughts about Astronomical Image Processing for Digital Cameras |
the scope of this article
I gained all my experience in the field of digital image processing for astronomy by dealing with raw frames of 16 bit resolution of luminance per pixel coming from a specialized astronomy CCD cameras.
These cameras typically do not have too many pixels.
One million pixels is already regarded to be much.
On the other hand a digital daylight camera like the Canon EOS 300D has more than 6 million pixels.
But when saving the frames as JPG they will only have 8 bit of luminance resolution per pixel.
On the other hand the image coming from an astronomical CCD camera is arriving at the PC totally raw and unchanged.
This is not true for many digital daylight cameras.
Another point is that the software of astronomical cameras is saving the files in FITS which is a lossless and uncompressed format all the astronomy software packages can read.
This is not the case for the uncompressed raw format of digital daylight cameras which are differing from manufacturer to manufacturer.
These differences can cause the need for a different approach in image processing to get the best possible results.
picking the 'correct' format for storage
A good astronomical photo is one which is sharp and with less noise as possible.
This is usually achieved by using many exposures of the same object.
These many frames must be selected, registered and then be overlayed to give the resulting image with a better signal to noise ratio.
Later on there is a big set of filters and treatments that will improve the image to show as much of a faint nebula or dim stars as possible.
This is not necessary for day light photos.
And that is the reason why the software bundled with the camera is not able to perform all theses tasks.
Unfortunately this software is usually the only one which is able to read the camera's specific raw frame format.
That is why picking the lossless and high resolution raw format is not very appropriate for most digital cameras.
However the JPG format can be used by many of the astronomical image processors.
That is why I prefer that format for the Canon EOS 300D.
Even the highest resolution of 3072 x 2048 in best JPG compression quality is downloading as fast as the 1300 x 1024 of the HX916 in uncompressed format.
That is very convenient and not wasting valuable exposure time at the telescope.
averaging or adding of the frames?
As we know from a television set the luminance is responsible for the overall sharpness of the image.
The luminance can be regarded to be the number of levels of gray in an image.
8 bit is equal to 256 levels of gray which is about the maximum the eye can distinguish.
That is why the JPG format has 8 bit of luminance (256 possible values of equal amounts of red / green / blue).
Unfortunately with astronomical photos there is the need to emphasize the darker parts of the image by histrogram stretching.
Preferably this is done in a logarithmic way and this will bring out the faint details of nebula and faint stars.
If you perform this on an 16 bit image there is no problem at all because there is as many as 65536 levels of gray.
The difference between these levels of gray is so small, that even a stretch by factor 10 is still lokking smooth to the human eye.
Not so for 8 bit resolution!
After the stretch you will clearly see 'steps' where there should be a smooth blending.
 
histogram stretch with 16 bit and 8 bit of luminance
Averaging 16 bit frames is giving 16 bit at the end and that is just perfect for further processing.
There is no practical difference of adding and averaging with 16 bit raw frames.
But averaging 8 bit frames is giving 8 bit at the end and that is not enough.
Therefore we have to gain more bits of luminance and this can be done by adding the pictures instead of averaging them.
# of frames
added | resolution in bit
of luminance |
| 1 | 8 |
| 2 | 9 |
| 4 | 10 |
| 8 | 11 |
| 16 | 12 |
| 32 | 13 |
| 64 | 14 |
| 128 | 15 |
| 256 | 16 |
While 256 frames might look like really much it is normal for me to add from 25 to 100 frames.
This is resulting in 13 to 15 bit and considerably better than 1 to 4!
dark frame subtraction
There is only one reliable way of subtracting a dark frame from a raw frame to increase signal to noise ratio:
Take the dark frame at the same temperature, the same exposure time and the leave it completely unprocessed.
A digital camera has no contoller to keep the temperature os the sensing chip constant.
Though taking the dark immediately after each raw frame and using these pairs of raw/dark is a good estimation.
But that means losing 50% of the possible exposure time and hence losing SNR!
Please refer to Thoughts about Image Calibration for low dark current, unregulated Amateur CCD Cameras to increase Signal-To-Noise Ratio (SNR) for more detail.
But the killer for using dark frames with the Canon EOS 10D or Canon EOS 300D is the fact that these cameras use an undocumented noise reduction algorithm.
Of course they would also use this for the dark not knowing that it shall be unprocessed to give reliable results.
picking the 'correct' exposure time and ISO setting
A longer exposure time per frame means better SNR but not necessarily a better image!
The first restriction for the longest possible exposure time is the quality of the guiding.
This is depending on the quality of the mount, the quality of the polar alignment (image rotation), the focal length of the optics (the smaller the longer), the resolution setting of the camera (the higher the shorter) and seeing effects of the atmosphere.
At my largest focal length of 1200 mm and the highest resolution I can usually expose for up to 300 seconds.
My setup is restricted here by the differential flexure between imaging optics and guide scope.
In case of bad seeing or very dim guid stars I have decrease that to 60 or 120 seconds.
Now, for that maximum value I try to find the proper ISO setting of the camera so that neither the brightest part of the nebula, galaxy or the core of a cluster is not over exposed.
The lower the ISO setting the lower the noise will be in that frame.
In the case of the imaging optics having a focal ratio of f/4 my ISO settings are varying from 200 to 1600 depending on the sky object.
That is - by the way - the reason why a beginner with less experience or a low quality mount should pick low focal lengths.
With a fast 50 mm photo lens you can lower the ISO rate and at the same time expose longer to create very good photos.
You just have to pick the larger sky objects :)
picking the 'correct' image processing settings within the camera
Digital cameras will process the images before storage.
This is fine for daylight scenes but may be not for astronomical photos.
Sharpening is very noise sensitive and hence should be done as one of the last steps, definitely after adding the single frames.
Avoid any sharpening and try to set the camera to neutral or soft.
Contrast and color saturation is a different story.
While these are not necessary for storage as raw files they can help a lot when saving as JPG.
The reason is that the raw format (the camera's internal picture format) has more bits per color.
That can be 12 or even 16 bit.
When saving as JPG the camera has to decrease that and hence is affecting luminance and color saturation.
Now in astronomy with all these faint objects the color saturation and contrast is very low anyway.
Of course we want to 'keep' as much as possible when the camera is reducing the number of bits.
|
|
|