Binary exponential fractal functions
binary exponential functionsfractal functionsHaar functionsmultiresolution analysispixel functionssignal and image compressionWalsh functions
Hausdorff and packing measures (28A78) Fractals (28A80) Approximation by other special function classes (41A30) Fourier series in special orthogonal functions (Legendre polynomials, Walsh functions, etc.) (42C10) Nontrigonometric harmonic analysis involving wavelets and other special systems (42C40) Computing methodologies for image processing (68U10) Image processing (compression, reconstruction, etc.) in information and communication theory (94A08)
In the problems of signal and image compression, one has to approximate very peculiar functions, far from continuous and resemble fractals. Probably this was the motivation of \textit{M. F. Barnsley} and \textit{L. P. Hurd} [``Fractal image compression (1993; Zbl 0796.68186)] to develop the so-called Iterated Function Systems (IFS) techniques, creating special instruments for approximation of every single object. The authors of this paper follow the same idea to adjust the instrument for approximation to the individual object to be approximated. They define a set of function \({\mathcal B}\), called Binary Exponential Functions (as the function \(e^x\) belongs to this set, too). An orthonormal system of functions, defined through transformations of Haar and Walsh type is considered, that may have arbitrary fractal dimension and allows fast Fourier transformations. In the preliminary sections, the authors introduce the basic notions of binary fractal functions, completed graphs, Hausdorff continuity, pixel functions with a prescribed resolution \(s\), starting functions, fractal dimension of functions of such classes, Walsh and Haar transformations, most of them initiated in \textit{Bl. Sendov} [``Hausdorff approximations, (1990; Zbl 0715.41001)]. They prove that for every binary fractal function \(f\) and every natural number \(s\), there exists a binary exponential pixel function \(B_s (f;x)\) with resolution \(s\) of best \(L_2\)-approximation to the function \(f\) (without claiming to be unique). In practice, for compression of signals and images, the authors work in an \(N=2^s\)-dimensional space of pixel functions \(f \in {\mathcal P}_s\). They propose the following procedure: 1) For every function \(f \in {\mathcal P}_s \subset {\mathcal F}\) (\({\mathcal F}\) supposed to be a rich space of functions), choose the function \(g \in {\mathcal B}\) which approximates the function \(f\) and depends on \(s\) parameters; 2) Expand the function \(f\) by an appropriate \(N\)-dimensional orthonormal system closed in \({\mathcal F}\) and find the coefficients vector \(F_N\), describing the signal (image); 3) If the entropy of the vector \(\|F_N\|^{-1} F_N\) is small, this transformation (system) is useful for compression. The authors present also computer experiments with real signals and compare the quality of their method with some classical methods used for compression. The theory in this paper may be explained in the terms of the multiresolution analysis [\textit{Bl. Sendov}, East J. Approx. 3, No 2, 225-239 (1997; Zbl 0894.42011)].
This page was built for publication: Binary exponential fractal functions
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4941483)
