Finite orthogonal transforms and multiresolution analyses on intervals (Q1361401): Difference between revisions

The important discrete orthogonal wavelet transform decomposes a biinfinite sequence of data into detail components associated with scales of different magnitudes. However, in practical applications usually only a finite set of data is available. The author introduces a new approach to overcome this problem. The method is based on periodic extensions of the data. The straightforward finite matrix formulation of the wavelet transform on the periodized data leads to non-banded matrices. The main idea in the paper is to make the wavelet transform essentially banded by using orthogonal matrix transformations. The resulting (non-unique) finite wavelet transforms are orthogonal, and they give rise to orthogonal multiresolution analyses on an interval. The properties of these multiresolution analyses are discussed in detail.