Quote:
Originally Posted by Peter Ward
A physicist has some washing that needs doing, and sees a sign in a shop shop-window down the road. "Laundry done here"
He hauls the washing into the shop. The shop-owner (by co-incidence named Albert) is a Mathematician. The Physicist asked when he can have his washing done?
"Oh no!" "We just make signs! "
The point of my tale?
While the math works, I'd suggest it ignores the elephant in the room.... ...atmospheric turbulence...
Turbulence spreading light beyond a pixels location is why this works. With perfect seeing it would fail, and you'd need to move the sensor (i.e. Drizzle)
Improving the S/N ratio by stacking images (assuming random noise) and getting multiple samples of a stellar location (effectively averaging its location to a precision less than a single pixel) makes the difference....
... but I still fail to see how, the application of a constant (i.e make pixel X bigger by 1.5) improves anything with floating point calculations that follow.
I suspect a 1.5 pixel Gaussian spread/blur applied prior to stacking would end up looking the same.
|
Peter the images are randomly dithered so all detail is sampled at a finer mesh than the pixel size of the sensor. It would be useful to rotate the sensor as well only to eliminate the factor of nearly root two along the diagonals of the pixels.
I just cannot see why you find this so difficult to understand.
If the seeing is perfect then this will give a resolution of about four seconds of arc rather than a bit over six seconds of arc. The best a single image with this optic and camera can do is 3.1 seconds of arc per pixel which is really a resolution of 6.2 seconds of arc.
All I am claiming is that at best my resolution is four seconds of arc for all practical puposes.
It is even worse than this even if seeing is about 2 seconds of arc. Long exposures and mount tracking errors of even 1 second of arc rms will add to this blurring.
All this means that it is not just the undersampling of the sensor that is the limit of the final resolution. By using this method real practical gains can be made in resolution. After all the best that can be done is limited by the seeing and guiding errors.
The PMX rms guiding errors are less than one second of arc. So on a night of typical seeing say two seconds of arc my system can image to a resolution limited by seeing and guiding rather than inherent sensor resolution.
My choice of 600mm focal length was quite calculated as I am not really limited by seeing, as even when working at best performance of the system the seeing has to be worse than three seconds of arc to affect my imaging resolution.
Where this is method of resolution enhancement is really useful is when binningx2 to get four times the sensitivity and well depth. Rather than doubling your sensor 'pixels' in size they only go up by a factor of root two.
Bert