Shift invariant spaces of multivariate anisotropic functions on the torus (Q2862339)

This well-written booklet coincides with the doctoral thesis finished 2013 at the University of Lübeck. As known, shift invariant spaces and multiscale decompositions of functions are powerful tools in signal and image processing. The author develops a framework for anisotropic periodic multiscale analysis with the help of related shift invariant spaces. Further, he presents anisotropic periodic wavelet algorithms for the decomposition and reconstruction of periodic \(d\)-variate functions on the pattern of grid points. Thus the author continues the investigation of \textit{D. Langemann} and \textit{J. Prestin} [Appl. Comput. Harmon. Anal. 28, No. 1, 46--66 (2010; Zbl 1185.42033)].NEWLINENEWLINE In this booklet, the author considers functions \(f:\, {\mathbb T}^d \to {\mathbb C}\), where \({\mathbb T}^d := {\mathbb R}^d/2\pi {\mathbb Z}^d\) is the \(d\)-dimensional torus. For a regular matrix \({\mathbf M}\in {\mathbb Z}^{d\times d}\), the lattice \(\Lambda({\mathbf M}) := {\mathbf M}^{-1}\,{\mathbb Z}^d\) is 1-periodic and the pattern \({\mathcal P}({\mathbf M})\) is defined by \(\Lambda({\mathbf M}) \cap [0,\,1)^d\). A discrete Fourier transform on the pattern \({\mathcal P}({\mathbf M})\) is introduced together with a corresponding fast algorithm. For details see also author's paper [Appl. Comput. Harmon. Anal. 35, No. 1, 39--51 (2013)].NEWLINENEWLINE A linear subspace \(V \subset L^2({\mathbb T}^d)\) is called \({\mathbf M}\)-shift invariant, if \(f\in V\) implies \(f(\cdot - 2\pi x) \in V\) for all \(x\in {\mathcal P}({\mathbf M})\). The properties of \({\mathbf M}\)-shift invariant subspaces are analyzed. Using \({\mathbf M}\)-shift invariant subspaces and discrete Fourier transforms on the pattern \({\mathcal P}{\mathbf M}\), the interpolation problem on \({\mathcal P}{\mathbf M}\) is discussed. Applying periodic Strang-Fix conditions, the interpolation error is estimated with respect to an anisotropic Sobolev norm.NEWLINENEWLINE The anisotropic periodic multiresolution analysis is defined by a nested sequence of \({\mathbf M}_j\)-shift invariant subspaces \(V_j \subset L^2({\mathbb T}^d)\) \((j=0,1,\dots)\), where \({\mathbf M_j} = {\mathbf J}_j {\mathbf M}_{j-1}\) with regular matrices \({\mathbf J}_j\in {\mathbb Z}^{d\times d}\) and the unit matrix \({\mathbf M}_0\). Using fast Fourier transforms on the pattern \({\mathcal P}({\mathbf M}_j)\), fast decomposition and reconstruction algorithms for \(d\)-variate anisotropic periodic wavelets are presented. Finally, the author introduces well-localized de la Vallée Poussin-like scaling functions and related wavelets. Numerical examples are given for \(d=2\).