| Image search |
To bin or not to bin?

Celestron Nightscape 10100 OSC Camera
Loadstar X2 Guide Camera with Off-Axis Guiding
Celestron CGEM Mount
8 |
The image of NGC7331 was taken on the 5th and 12th September. The combination of the 2,321mm focal length SCT at f/10 with the Celestron Nightscape 10100 camera has an image scale of 0.42 arc seconds per pixel. The resolution limit of the system, applying the Nyquist sampling principle, will be 0.84 arcseconds. Given the seeing in Cornwall is usually between 1.0-2.5 arc seconds limited, I am probably slightly oversampling using the camera un-binned.
To capture this image I experimented with 2x2-binning the camera. With an image scale of 0.84 arc seconds per pixel and a resolution limit of 1.68 arcseconds, I was actually slightly under sampling the image acquisition (seeing limited resolution was 1.0 and 1.5 arc seconds on the 5th and 12th September respectively). The 2x2-binned images were captured as a total of 34x600-second light frames.
I used drizzle integration in PixInsight, with a scaling factor of 2x, to compensate for the under sampling.
I have analysed the results of: a 1x1 un-binned integrated image; a 2x2-binned integrated image; and a 2x2 drizzle-integrated image. Interestingly, the SNR of the 2x2-binned integrated image was higher than the 1x1 unbanned integrated image, as expected; and the 2x2-binned drizzle integrated image had a further significantly improved SNR over the 2x2-binned integrated (without drizzle) image. The analysis was carried out using the PI sub-frame selector.
The significant improvement of SNR of using 2x2-binned drizzle integration rather than 1x1-un-binned integration, appears to also be reflected in an improvement in resolution of structure within the NGC7331 image. Obviously no process tool can resolve information that does not exist on the CCD in the first place, but the technique does seems to have resolved information that was not so easily extracted in the usual 1x1 un-binned process.
Comments
1x1 and over-sampling
Within reason, over-sampling is beneficial in that if you have a good estimate of the total PSF (a combination of the transfer function of the optical train and of external causes such as seeing, tracking errors and so forth) and if the SNR is high enough, deconvolution will recover structural details which are not immediately apparent.
I suggest that you try to create the PSF for your 1x1 binned image, possibly by modelling the bright stars as point sources, and seeing what you can discover through deconvolution. Ground truth should be available from HST observations. You may wish to process an enlarged version of your original data (bicubic spline interpolation perhaps) to see whether sub-pixel scale processing is useful.