Fri Mar 16, 2018 12:33 pm
Generally it is better to avoid transforming if possible. It depends on the shot ofcourse, change in perspective and shape of surface make things more difficult.
The best way is to clone straight from the image with no transformation. This means using rounded pixel transforms to not introduce any blurring. Works great with grain. But it doesn't work that well when source and target change perspective relative to each other. For ground this might not be the case though, so I'd suggest cloning from nearby areas without any transforms except moving/scaling the clone source and target areas.
If previous solution doesn't work out, second option is to transform the plate so that you do cloning on stabilized plate. But in the end only re-transform the cloned area and merge over original background. You may need to sharpen the cloned patch and/or reintroduce grain. Never introduce transform filtering to whole plate if you only worked on part of it.
There is also a hybrid way where you generate patch on transformed source but patch does not contain image itself, instead you feed it with UV (or ST or however someone names them) coordinates from original image. After re-transforming the patch you use the modified UV-s for lookup into original image. This is a bit contorted approach, but might prevent degradating the image too much in some more elaborate cases.
I do stuff.