An anisotropic directional subdivision and multiresolution scheme (Q2355186)

Multiwavelets are a new addition to the body of wavelet theory. Application of multiwavelets is a fast developing field and the literature on the subject is rapidly growing. Virtually all multivariate data are typically dominated by anisotropic features such as singularities. In order to handle directional singularities, standard wavelet approaches were extended to the concept of discrete shearlets by a group of researchers [\textit{G. Kutyniok} and \textit{D. Labate}, in: Shearlets. Multiscale analysis for multivariate data. Boston, MA: Birkhäuser. 1--38 (2012; Zbl 1251.42010)]. According to the authors, this extension, however, suffers from some drawbacks. Basically, extension from wavelets to multiwavelets, uses scaling matrices of large determinant, giving rise to complexity issues. The paper presents a different approach, based on a matrix with small determinant. The authors' main contribution is motivated by a very recent extension based on the concept of multiple subdivision using different scaling matrices and filters in each step of the subdivision process. Such a strategy, as claimed by the authors, allows directionally adapted processing of the data. The discrete Shearlet transform (DST) introduced in [loc. cit.] has a disadvantage of relatively large determinant of the matrices involved. The paper presents a different approach, based on multiple subdivision schemes with matrices having small determinant resulting in the reduction of complexity substantially. A key feature of DST ``slope resolution'' is also maintained; a useful fact as shown by the authors employing the tools of set theory, group theory and toplology. Finally, as a special case, the authors present an explicit construction of bivariate filter bank with an example of application highlighting the advantages of multiple scheme. The authors conclude the paper with multiwavelet processing of an multiresolution image, frequently obtaining performance superior to the comparable scalar (univariate) wavelet transform.