Performs largrange polynomial bicubic interpolation on two-dimensional matrices. This project is specifically tailored for the interpolation of custom W-Projection convolutional kernels for a currently private research project.
The requirements for this interpolation process are to scale from some two-dimensional source matrix, down to some two-dimensional destination matrix. It assumes the dimensions of both matrices are some power of two, and that during the interpolation process, the entire first row and colum of the source matrix are not included during interpolation. This is needed as the first row and column contain redundant information obtained during the FFT of the source matrix during the creation of kernels.