Any bad pixels in the input NDF may be replaced by a constant value. Input NDFs need neither be square, nor be a power of 2 in size in either dimension; their shape is arbitrary.
The Hermitian transform is a single image in which each quadrant consisting of a linear combination of the real and imaginary parts of the transform. This form is useful if you just want to multiply the Fourier transform by some known purely real mask and then invert it to get a filtered image. However, if you want to multiply the Fourier transform by a complex mask (e.g. the Fourier transform of another NDF), or do any other operation involving combining complex values, then the Hermitian NDF must be untangled into separate real and imaginary parts.
There is an option to swap the quadrants of the input NDF around before performing a forward FFT. This is useful if you want to perform convolutions with the FFTs, since the point-spread function (PSF) image can be created with the PSF centre at the array centre, rather than at pixel (1, 1) as is usually required.
KAPPA --- Kernel Application Package