Computational cost

For a data cube with dimensions $n=n1 \times n2 \times n3$, each iteration requires n1 forward and reverse 2D FFT's. Therefore, number of operations per iteration is about $8 n (1+log(4 n2 \times n3))$. The number of iterations is a function of the variability of the structure and the degree of weighting. For instance, if the structure is constant with depth, then it will be flattened in one iteration. On the other hand, if a weight is applied and the structure changes much with depth, it may take as many as 100 iterations.