Closed kolhizin closed 5 years ago
1&2 are memory-heavy 3 relies on localization
Implemented decreasing window algorithm, i.e. 1) select window twice as large as image 2) calculate offset for each point based on this window (leading to same offset for all) 3) shrink the window twice 4) use new window and previously calculated offsets as initial point 5) repeat
Ideas: 1) Finite-element method 2) Analog of Fourier transform 3) Calculate distance in both directions and approximate with function of minimal variance