Superscaling

[BrendanC]

Hi,

I've recently started printing some of my efforts on canvas, and to do this I've had to scale them up. I've tried using combinations of AI upscalers, Photoshop, Denoise and so on, and managed to get something OK.

However, given the superb signal processing skills that drive StarTools, I was wondering whether a superscaler module could be considered? I totally understand that ideally one would do the exact opposite ie binning to increase the SNR, but it would be a nice option to have, when scaling images up to print size.

Thanks
Brendan

[admin]

Hi Brendan,

Upscaling should be a pretty simple affair. It should definitely not involve (fake) detail-manufacturing algorithms, and it should certainly not require any further denoising. A simple bilinear or bi-cubic upscaler should suffice.

If you had to bin due to oversampling, then you would not have lost any detail anyway, so upscaling would simply yield that same (or more - due to decon) detail that was contained in your image at its original unbinned size.

You can try this with any oversampled dataset yourself. Load it , stretch it, save that away, now bin it (making sure it is not oversampled), upscale that GIMP or PS and you should have an image that is virtually identical to the unbinned image you saved away earlier.