Total Variation Inpainting for image scaling

[LigH]

I just heard in another forum about the existence of NUKE by The Foundry (https://www.thefoundry.co.uk/products/nuke/), which seems to be a comprehensive video editing framework. This also implements an upscaling technique called TVIScale (http://help.thefoundry.co.uk/nuke/8.0/content/reference_guide/transform_nodes/tviscale.html). To me it sounds like "an educated guess about probable details". I can find a few papers mentioning a Split Bregman algorithm and Wavelets. Is there anyone with a little more insight?

[luquinhas0021]

Total variational impainting is a technic that try to recover the miss information in image/frame. How? Suppose you have an image plenty of noise, but at least one edge of each segment of each object is correct represented. The TVI decompose the image in two layers, put of noise and try reconstruct the miss edges taking the information of close edges. For that, a lot of techniques can be used, such Taylor Series, Fourier Series, Shearlets, etc. Search for "Shearlet" on Google, enter in Shearlet official page and research for it applications.
I guess TVI scale in Nuke first scales the image using a sophisticated resizing algorithm, and then apply the TV filter.
Ask about comercial softwares in this forum doesn't work, because the most of users is so addicted in Avisynth, that's why don't know (Or don't like) about commercial solutions.

[LigH]

I just read (http://forum.videohelp.com/threads/373780-Using-Nuke-to-Upscale?p=2407137&viewfull=1#post2407137) that it may be impressive for single images, but in videos it may have issues with "temporal fluctuations", constructing missing parts slightly differently from frame to frame... a plausible warning.

[luquinhas0021]

Yes, and it must cause temporal aliasing/temporal noise, or not. A good implementation is:
first: resample
second: apply a temporal subpixel motion filter, called of super-resolution filter too.
finally; apply TVI
The super-resolution filter must "regularize" the video close frames behave, and take off some artifacts.