Spatially varying two-dimensional filter banks

In this paper we show how to construct spatially-varying filter bank decompositions. The idea is that we can use different filters over different portions of a two-dimensional signal, or even completely vary the structure of the analysis and synthesis trees. To do so we associate with the filters to be used, a set of boundary filters, which are applied whenever a transition is to be made. These boundary filters allow the change to made without distortion or redundancy.