Looking at cv::remap, I can use that to de-fisheye an image I have with a translation table that I have prepared earlier for a particular model of lens, and then use cv::warpAffine to correct for camera roll, however I suspect that there will be a significant cumulative loss of quality doing the two operations sequentially versus combining both operations. Unfortunately, my maths aren't up to working out how to preprocess the maps, so can anyone help?
(Note, I require an OpenCV 2.x solution, not 3.x)
