Directional Haar wavelet frames on triangles

DOI10.1016/J.ACHA.2009.03.002MaRDI QIDQ837545zbMATH OpenOpenAlexFDO

Publication date 20 August 2009

Published in Applied and Computational Harmonic Analysis (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1016/j.acha.2009.03.002

composite dilations dual frames Haar wavelet frames image denoising nonseparable wavelets sparse representation

Nontrigonometric harmonic analysis involving wavelets and other special systems (42C40) Image processing (compression, reconstruction, etc.) in information and communication theory (94A08)

Review text

The composite wavelets such as those considered by \textit{K. Guo} and \textit{D. Labate} [SIAM J. Math. Anal. 39, No. 1, 298--318 (2007; Zbl 1197.42017)] and by \textit{I. A. Krishtal} et al. [J. Geom. Anal. 17, No. 1, 87--96 (2007; Zbl 1124.42026)] are modified in one sense and extended in another to produce Haar wavelets on triangles. They are wavelets with respect to composite dilations with expanding matrix \(A=2I\) and the dihedral group \(D_8\) of isometries of the square \([-1,1]^2\), considered as a group of shear matrices. The geometry works as follows: \([0,1]^2\) is subdivided into four triangles, each with a vertex at \((0,0)\) such that each triangle is obtained from the other under one of the shear matrices. The triangle with vertices \(T_0\sim\{(0,0), (1,0), (1,1/2)\}\) and the triangle \(T_1\) with vertex set \(\{(0,0), (1,1/2) , (1,1)\}\) generate a subdivision of \([-1,1]^2\) consisting of 16 triangles under the action of \(D_8\). The characteristic functions of these triangles are thought of as the basic scaling functions, and it is shown how to subdivide \(T_0\) and \(T_1\) each into 4 further triangles such that the composite dilations generated by \(A=2I\) and \(D_8\) generate a ``self-affine tiling of the plane under integer translations. The subdivision gives rise to a multiresolution structure and thus to a tight frame of ``directional Haar wavelets supported on the subdivision triangles with values \(\pm 1\). They can be regarded as having an orientation defined by their center of mass. It is proved that the wavelet system forms a Parseval frame for \(L^2(\mathbb{R})\). Filter bank and wavelet transform algorithms are described. Compression and denoising algorithms that take into account directionality are described and numerical comparisons are made with \textit{E. J. Candès} and \textit{D. L. Donoho}'s curvelets [Appl. Comput. Harmon. Anal. 19, No. 2, 198--222 (2005; Zbl 1086.42023)] and Do and Vetterli's contourlets [\textit{M. N. Do} and \textit{M. Vetterli}, Contourlets. Welland, Grant V. (ed.), Beyond wavelets. San Diego, CA: Academic Press/Elsevier Science. Studies in Computational Mathematics 10, 83--105 (2003; Zbl 1086.65125)].

Recommendations

Cites work

Cited in

(12)

This page was built for publication: Directional Haar wavelet frames on triangles

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q837545)