Legacy ImageMagick Discussions Archive

Found a reference: http://www.mrsc.ucsf.edu/~plarson/pubs/ ... -rBURS.pdf They (P. E. Larson and J. M. Pauly) tried Kaiser and Hamming windowing. Within their application, Kaiser did better.

Won't have time to look into this for a while, but I would bet that if EWA benefits from a different window than Jinc, it would be a Kaiser window with tuned alpha (and, possibly, "(de)blur").

For example, I bet you anything that I can find a reasonable alpha/(de)blur combo that preserves exactly vertical/horizontal lines under no-op. The parameter count is right if I limit myself to extents, for 3-lobes, of 3 or less.

Matching numbers of equations and unknowns does not guarantee a solution, but it's a reasonable basis for wishful thinking.

I think that the smaller the number of lobes, the more a filter is likely to benefit from a careful choice of window function. (Lanczos2, we're coming for you!)