Non-separable wavelet

Content sourced from Wikipedia, licensed under CC BY-SA 3.0.

Non-separable wavelets are multi-dimensional wavelets that cannot be written as a simple tensor product of one-dimensional wavelets. They have been studied since 1992. They offer advantages, including more design options that can lead to better filters. In multiple dimensions, sampling uses lattices (such as the quincunx lattice) rather than a regular grid. The filters themselves can be separable or non-separable, independent of the sampling lattice. So, non-separable wavelets can sometimes be implemented in a separable way. Unlike separable wavelets, non-separable wavelets can detect structures in directions beyond horizontal, vertical, or diagonal, reducing directional bias.