Multifilters with and without prefilters (Q698562)

Multiwavelets constitute a useful generalization of the classical wavelet theory, since they can possess several desirable properties, like orthonormality and symmetry at the same time. Unfortunately, a high approximation order \(p\) of the multiscaling function does not imply that the corresponding low-pass multifilter preserves the discrete time polynomials up to the degree \(p-1\). The authors deal with two methods to overcome this problem: the more usual method consists in the incorporation of prefilters. The authors describe an explicit technique for the construction of such prefilters using matrix factorization techniques. However, prefilters which explore the full approximation order of multifilters are in general of a very complicated form. Therefore the authors further introduce a class of multifilters (called ``full rank multifilters'' by the authors) which allow to obtain maximal approximation order without prefiltering. The construction of these multifilters is based on the lifting scheme. Image compression examples demonstrate the performance of some full rank multifilters.