On Fri, Nov 12, 2021 at 5:33 PM Terry Pinfold <[email protected]> wrote:
> I have found this publication which may give some insight into the > sharpening techniques used for hubble images. I suspect some of the > developers can understand the complex formulas and weave their magic. > > > https://www.researchgate.net/publication/252709898_APEX_blind_deconvolution_of_color_Hubble_space_telescope_imagery_and_other_astronomical_data > As Chris said, Aurélien Pierre worked on that, but with time he shifted to a different algorithm ( https://discuss.pixls.us/t/got-an-image-problem-go-see-the-image-doctor/14518). Blind deconvolution on general images with an arbitrary camera+lens combination is awfully slow for the benefit it provides. I understand even the "Image doctor" version described on the link (not based on blind deconvolution) was not performant enough, and that led him to develop the "diffuse or sharpen module", based on diffusion theory. > BTW, I have just looked at the new diffuse or sharpen module. It is not > for the faint hearted, but I will explore it in more detail as it showed > good promise on a very challenging image I tested it on. The preset was way > too heavy for the image but reading the user guide I started getting some > better results. Is the diffuse and sharpen module based on wavelets? I am > trying to get my head around the meaning of the four orders. > You may get some pointers and math background for the PR in github ( https://github.com/darktable-org/darktable/pull/7669) and the related discussion on Pixls.us ( https://discuss.pixls.us/t/diffuse-module-is-now-in-the-master/25765/56), besides obviously the documentation for the next version ( https://darktable-org.github.io/dtdocs/en/module-reference/processing-modules/diffuse/ ). The math behind follows the theory of physical diffusion, either 'forward in time' to diffuse or 'backwards' to un-diffuse or sharpen. This means it has to solve a (quite big) system of coupled second order partial differential equations. To do that in a reasonable time moves to Fourier space and uses wavelets to work at multiple scales, but it's not 'based on wavelets' per se. Best regards, Guillermo ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to [email protected]
