| article: Resolution of Bayer Matrix encoded CCD cameras in Astronomy |
What is the Bayer Matrix?
The Bayer Matrix is a patented way to create one-shot colour cameras.
With these cameras there is no need to place red, green and blue filters in front of the whole camera and recombine these frames later to form a colour image.
Instead each pixel sensor of the CCD sensor has a tiny little colour filter in front of it.
Four pixels are grouped together to form a super pixel - that is why we can call it a 2x2 matrix.
Each of these superpixels has 2 green pixels, one red and one green.
Bayer Matrix decoding
To form a colour image the information about each colour channel in one super pixel must now be decoded.
The easiest way to do this is to regard one superpixel as one final pixel, take it's red information from red, the blue from blue and the green from the average of both greens.
The result will be a decoded image with 1/4 of the pixels.
On the other hand the geometrical information of that decoded image will be 100% correct and there will be no colour artefacts.
In addition there will be less noise and a bigger full well depth of each superpixel.
The latter is of course strongly depending on the wavelength of the object to be photographed.
It will be between 4 times for white sources, 2 times for green and no increase for red and blue sources.
Most of the decoders are outputting the full number of pixels (full resolution) in the decoded image.
There are many different approches to do so.
A simple way is to interpolate the missing colour by its neighbours.
For example in the red-filtered pixel its blue and green value must be retrieved from the neighbours.
In other words there will end up information in this red pixel which is taken from different places around that pixel.
This is even decreasing the resolution compared to the simple approach described above!
Of course there are more sophisticated approaches to do the job.
But they all are fighting hard with colour artefacts at the corners and geometrical distortion of the final image.
The best way might be to separate the luminance from the colour and increase the resolution for luminance only.
This way there is a chance for about 40% better resolution compared to the superpixel algorithm.
The true resolution is hence still lower than the ouput image might look like and the result must hence be spatial distortion in the image.
Conclusion
That's is why I suggest to count the number of pixels in super pixels (1/4 of the pixels on chip) and hence the pixel area four times as big.
It seems to be poor to lose so much resolution.
But first of all this reduction in resolution means a better signal to noise for each super pixel compared to a single pixel.
And second there is a good chance to use super resolution when combining many frames of the same image to the final frame.
This is due to the fact that during registration and stacking there is a way to calculate the centroid of a star to sub-pixel precision.
In other words: in astronomy you can easily gain back some resolution for your final frame!
I guess 50 to 100% more resolution should work fine.
Click here for more detial.
Very tiny pixels in existing Bayer Matrix cameras
Knowing all that did change my attitude about the Starlight SXV-MX8C camera completley!
This camera has a pixel size of 3.125 mym only.
By the Nyquist theorem this camera can properly sample spot sizes of about 9 mym (2 times the diagonal of a pixel).
But the spot size of my coma corrector is 12 mym at its optimum hence the MX8C would over sample the telescope by 50% at least.
But taking the superpixels into account the MX8C will be just perfect!
From my very personel perspective the MX8C is a perfect 1 Megapixel one-shot colour artefact and geometrical distortion free camera.
Let's wait for the first results...
For a sortable comparison chart of the basic data for these CCD cameras click here.
Click here to see how this (and many other) camera fit to your telescope's focal length.
|
|
|
