Where/when to downsample a drizzle inetgration

[VoidPointer]

I have a rather big data set that I acquired with a one-shot color camera and a 2x drizzle integraion yields a pretty good linear image to process further.
While having a lot of resolution is obviously never a bad thing, the resulting image is quite large and I am tempted to downsample it before applying deconvolution and some noise-reduction. My machine is not slow but optimizing deconvolution settings (where deringing in a preview can behave differently then in the full image) on a 50MP image is starting to feel a bit slow.
I won't need the full resolution data in the end so I'm fine with downsampling at some point. Ideally as soon as possible in the workflow to save some time when working with processes like deconvolution that require quite a bit experimenting to yield optimal results

Will downsampling the linear image before deconvolution reduce the effectiveness of deconvolution?
I.e. is there a significant difference to be expected between:
a) downsample -> deconvolution ->noise-reduction vs.
b) deconvolution -> noise-reduction-> downsample ?

Cheers,
Lars

 

[Juan Conejero]

Hi Lars,

Will downsampling the linear image before deconvolution reduce the effectiveness of deconvolution?

If your data are undersampled (and hence drizzle is necessary), then yes. One of the benefits of drizzle is that the PSF of the image is much better represented, and hence can be fitted with higher accuracy. A more accurate PSF model used for deconvolution will always give better results. This can also be true even for well sampled or slightly oversampled images.

 

[VoidPointer]

Thanks for the clear explaination. At least, now I know it should be worth the wait, working with the full data...

On the bright side, figuring out the "perfect" deconvolution paramters for an image has become almost a form of meditation for me and the more attention one pays to getting that right the greater the reward in the end ;-)

Cheers!
Lars