…I prefer to round this ‘ideal N’ up to the next highest even number, and declare this to be N…
…It is resized without interpolation and periodic boundary policy…
Better to round to the nearest than rounding up. This was adding two: N=s+2 instead of N=s for square images of dimension s. Changed boundary policy to be Dirichlet (off-edge pixels are declared to be black). That doesn’t inject repeated coefficients. Seem to have more reasonable vector rendering now…
My diagonalizer is a memory pig though. Still. Time to try @David_Tschumperle new toys…