Hilbert and Riesz transforms using atomic function for quaternionic phase computation (Q2443910): Difference between revisions

The aim of the paper is to highlight the role of an atomic function (AF), \(up(x)\), as a kernel for the Hilbert and Riesz transforms for quaternionic phase computation and to show the advantages of using atomic functions over the global-phase approach which is based on the Fourier transform. Atomic functions are compactly supported infinitely differentiable solutions of differential functional equations with a shifted argument, i.e., \[ Lf(x)=\lambda \sum_{k=1}^M c(k) f(ax-b(k)),\quad |a|>1, \] where \[ L=\frac{d^n}{dx^n}+a_1 \frac{d^{n-1}}{dx^{n-1}}+\cdots +a_n , \] is a linear differential operator with constant coefficients. The atomic function \(up(x)\) considered in this article is defined by \[ up(x)=\frac{1}{2\pi}\int_{-\infty}^\infty \prod_{k=1}^\infty \frac{\sin (\nu 2^{-k})}{\nu 2^{-k}} e^{i\nu x}d\nu , \] and is extended to several variables by tensor product, e.g. \(up(x,y)=up(x)up(y)\). It follows from the definition of \(up(x)\) that the derivative of \(up(x)\) satisfies \[ dup(x)=2up(2x+1) -2up(2x-1). \] Another function used is the so-called Plop function, \(\operatorname{Plop}(x,y)\), which is a radial infinitely differentiable function with compact support. For one-dimensional signals the phase computation is carried out using the analytic part \(f_A\) of the signal \(f\) given by \[ f_A\left(x\right) =f(x)+ i f_H (x) , \] where \(f_H\) is the Hilbert transform of \(f\). For two-dimensional signals the phase is computed using the monogenic part \(f_M\) of the signal which is given by \[ f_M({\mathbf x})=f({\mathbf x}) +(i,j)f({\mathbf x})* \frac{{\mathbf x}}{2\pi |{\mathbf x}|^3 }, \] where * stands for convolution. The main result is to show that both the Hilbert transform and the partial Hilbert transform can be expressed in terms of the \(up(x)\) function and its derivative \(dup(x)\) and that the Riesz transform for two-dimensional signals can be expressed in terms of the Plop function. The authors then use these results to compute the quaternion analytic signals and the monogenic signals for local phase computation and discuss the advantages of their approach.