A new approach to the design of multidimensional non-separable two-channel orthonormal filterbanks

We propose a novel approach to the design of multidimensional non-separable two-channel orthonormal filterbanks and wavelets, which has been a challenging problem. We perform spectral factorization of the multidimensional product filter by jointly optimizing the product filter and the lowpass filter. The optimization is performed using iterative quadratic programming. Examples show that the resulting filterbanks have negligible reconstruction error and hence can be treated as orthonormal filterbanks and wavelets in practice