Oversampling?

[BlueAstra]

If I use my CCD camera on the scope at 1x1 binning my pixel FOV is 0.43 arcsec. I appreciate that this is probably below the seeing limit in my windswept rainy part of the country, but if I go to 2x2 / 0.86 arcsec that would lower the resolution (I'm ignoring the sensitivity increase for now). So my question is, given typical seeing in the UK, would the apparent resolution of the 1x1 0.4 arcsec + seeing limit be close to apparent resolution of 2x2 0.86 arcec + (less seeing effect?)? Is more resolution better regardless of seeing?

[thanders]

Theoretically you need at least to samples per unit of resolution. In Sweden seeing is normally no better than 2 seconds of arc. An optimal pixel scale should then be in the range of say 0.5-1.0 arc seconds per pixel. Oversampling will be of no good. Resolution can however be improved by some techniques like drizzling.

/T

[dph1nm]

Unless you are talking about planetary imaging, where you hope to capture moments of excellent seeing, then for the UK  0.86" is probably about right, or even a little small.  0.43 will almost certainly be oversampled.

NigelM

[BlueAstra]

What will be the effect of over sampling in seeing limited environment ?

[thanders]

Fewer photons per pixel meaning longer exposures, more noise.

/T

[swag72]

Theory, theory and more theory - Pah! 

I was imaging with a C9.25 last year at 0.32" and nothing bad seemed to be happening!!! I did an experiment and did some subs at 1x1 bin and some at 2x2 bin. When they were stacked I felt that the 1x1 bin had the edge they were a little sharper all round.

Now I have a different scope and am imaging at 0.47" - I always use 1x1 bin and do long subs. Of course it's easier with a faster scope and a shorter focal length, but I'd like to think that my images show that over sampling is not the killer that we are lead to believe. I am one for trying things and flying in the face of theories

[Martin Meredith]

From an information point of view undersampling is the real killer rather than oversampling. The latter is, if anything, just wasteful  .

But I'm not surprised that you found the oversampled versions sharper since some processes benefit from having more pixels, even if there is no new data in them. This is often due to the way that algorithms are implemented. An example: edge artefacts in any process involving convolution will be slightly reduced if the data is over-sampled.

Another possibility is that in moments of better seeing the effective FWHM is lower, and by not binning you are getting the benefits. This is entirely predicted by theory   .

Martin

[dph1nm]

You get more read-noise per unit area on the sky by observing 1x1 rather than on-chip binning 2x2 - whether that makes much difference to the final image depends on how much read noise you have in the first place (relative to other sources of noise).  I wouldn't worry about the unbinned pixels having fewer photons in them - you can recover this by binning afterwards in software.

NigelM

[thanders]

Theory, theory and more theory - Pah! 

I was imaging with a C9.25 last year at 0.32" and nothing bad seemed to be happening!!! I did an experiment and did some subs at 1x1 bin and some at 2x2 bin. When they were stacked I felt that the 1x1 bin had the edge they were a little sharper all round.

Now I have a different scope and am imaging at 0.47" - I always use 1x1 bin and do long subs. Of course it's easier with a faster scope and a shorter focal length, but I'd like to think that my images show that over sampling is not the killer that we are lead to believe. I am one for trying things and flying in the face of theories

Theory and praxis  . We are talking ball park figures. 0.32" or 0.47" is no big difference.

From an information point of view undersampling is the real killer rather than oversampling. The latter is, if anything, just wasteful  .

But I'm not surprised that you found the oversampled versions sharper since some processes benefit from having more pixels, even if there is no new data in them. This is often due to the way that algorithms are implemented. An example: edge artefacts in any process involving convolution will be slightly reduced if the data is over-sampled.

Another possibility is that in moments of better seeing the effective FWHM is lower, and by not binning you are getting the benefits. This is entirely predicted by theory   .

Martin

Agree on undersampling.

Oversampling gives no extra information. To avoid computational artefacts you might as well double the resolution in software first. Or?

You get more read-noise per unit area on the sky by observing 1x1 rather than on-chip binning 2x2 - whether that makes much difference to the final image depends on how much read noise you have in the first place (relative to other sources of noise).  I wouldn't worry about the unbinned pixels having fewer photons in them - you can recover this by binning afterwards in software.

NigelM

Signal to noise ratio is the interesting thing here. Why do we have on chip binning?

/T

[dph1nm]

Signal to noise ratio is the interesting thing here. Why do we have on chip binning?

As far as I can see, the only real scientific purpose to on-chip binning is to lower the overall read-noise contribution. It has practical benefits like fewer pixels to store and process of course.

NIgelM

[BlueAstra]

OK, I'll stick with 1x1

[ollypenrice]

OK, I'll stick with 1x1

I wouldn't. I'd experiment. If I'd have listened to every persuasive theory expounded on the net I wouldn't be using the capture procedures I now use and my images wouldn't be as good. (That's not to say they don't have a long way to go!!  ) I've read a gazillion proofs that thirty minute subs won't beat fifteen minute subs. I've got a gazillion subs that prove that they do.

Tim posted a while back saying much what Sara is saying, that even from Cventry he could extract more resolution from unlikely sounding pixel scales than was generally accepted.

I spent two years at 0.6 arcseconds per pixel against my wishes because the camera of the day wouldn't bin.  I'm not a fine detail junkie. What I like to have above all is signal. When I have that, I can make a picture. When I don't, I cant. So another factor is, 'What are you looking for?' and 'What do you enjoy in imaging?'

This isn't just about the numbers.

Olly