Classical Shearlet system forms a Parseval frame.

138 Views Asked by At

I have been reading about shearlets & frames and there is a part that I'd like help with. I have only studied the basics of functional analysis and wavelets.

First few definitions:

Classical shearlets

Let $\psi \in L^2(\mathbb{R}^2)$ be defined the following way $$\hat{\psi}(\xi) = \hat{\psi}(\xi_1,\xi_2) = \hat{\psi}_1(\xi_1)\hat{\psi}_2\Big(\frac{\xi_2}{\xi_1}\Big)$$ where $\psi_1 \in L^2(\mathbb{R})$ is discrete wavelet in the sense that it satisfies the discrete Calderòn condition, given by \begin{equation}\label{10} \sum_{j\in\mathbb{Z}}|\hat{\psi}_1(2^{-j}\xi)|^2 = 1 \quad a.e \texttt{ } \xi\in\mathbb{R} \end{equation} with $\hat{\psi}_1\in C^\infty(\mathbb{R})$ and supp($\hat{\psi}_1$) $\subseteq [-\frac{1}{2}, -\frac{1}{16}] \cup [\frac{1}{16}, \frac{1}{2}]$, and $\psi_2 \in L^2(\mathbb{R})$ is a bump function in the sense that \begin{equation}\label{11} \sum_{k=-1}^1 |\hat{\psi}_2(\xi+k)|^2 = 1 \quad a.e \texttt{ } \xi \in [-1, 1] \end{equation} satisfying $\hat{\psi}_2 \in C^\infty(\mathbb{R})$ and supp($\hat{\psi}_2$) $\subseteq [-1, 1]$. Then $\psi$ is called a classical shearlet.

Shearlet system

$$\text{SH}(\psi) = \{\psi_{j,k,m} = 2^{-\frac{3}{4}j}\psi(S_kA_{2^j} \cdot -m) : j,k \in \mathbb{Z}, m\in\mathbb{Z}^2\}$$

With these we have the following proposition:

Let $\psi \in L^2(\mathbb{R}^2)$ be a classical shearlet. Then SH($\psi$) is a Parseval frame for $L^2(\mathbb{R}^2)$.

Proof. \begin{align*} \sum_{j\geq 0} \sum_{|k| \leq \lceil 2^{j/2} \rceil} |\hat{\psi}(S^\text{T}_{-k} A_{2^{-j}}\xi)|^2 &= \sum_{j\geq 0} \sum_{|k| \leq \lceil 2^{j/2} \rceil} |\hat{\psi}_1(2^{-j}\xi_1)|^2 |\hat{\psi}_2(2^{j/2}\frac{\xi_2}{\xi_1}-k)|^2\\ &= \sum_{j\geq 0} |\hat{\psi}_1(2^{-j}\xi_1)|^2 \sum_{|k| \leq \lceil 2^{j/2} \rceil} |\hat{\psi}_2(2^{j/2}\frac{\xi_2}{\xi_1}+k)|^2 = 1 \end{align*} The claim follows immediately from this observation and the fact that supp $\hat{\psi} \subset [-\frac{1}{2}, \frac{1}{2}]$. $\square$

The problem that I have is with the last sentence of the proof. Does the claim follow from direct computation and the definition of frames or is there something more?

Thank you in advance.

1

There are 1 best solutions below

0
On BEST ANSWER

It does not follow immediately, but there are proofs in wavelettheory, which are quite similiar to this one.

I wrote a complete proof in my Bachelor thesis. What you need is the following result about the Fourier transform of a classical shearlet:

Let $\psi\in L^2(\mathbb{R}^2)$ be a classical shearlet and SH($\psi$) be the associated discrete shearlet system. Then the Fourier transfom of an element $\psi_{j,k,m}$ of SH($\psi$) ($j,k\in\mathbb{Z}, m\in\mathbb{Z}^2$) has the following form: \begin{align} \nonumber \hat{\psi}_{j,k,m}(\xi)=e^{-2\pi i\langle m,{(M_{j,k}^{-1})}^{T}\xi\rangle}\hat{\psi}({(M_{j,k}^{-1})}^T\xi)\quad\text{for }\xi\in\mathbb{R}^2, \end{align} where \begin{align} M_{j,k}=S_kA_{2^j}=\begin{pmatrix}2^j & 2^{j/2}k\\0 & 2^{j/2}\end{pmatrix}. \end{align}

Armed with this Lemma, we can proof the theorem:

Let $f\in L^2(\mathbb{R}^2)$ by the Plancherel theorem it holds \begin{align} \nonumber \sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\sum_{m\in\mathbb{Z}^2}\,\vert\langle f,\psi_{j,k,m}\rangle\vert^2=\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\sum_{m\in\mathbb{Z}^2}\,|\langle \hat{f},\hat{\psi}_{j,k,m}\rangle\vert^2. \end{align} And by the previous lemma, the Transformationssatz and the fact that supp$\hat{\psi}\subset[-\frac{1}{2},\frac{1}{2}]^2$ we can conclude that \begin{align} \nonumber \sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\sum_{m\in\mathbb{Z}^2}\,|\langle \hat{f},\hat{\psi}_{j,k,m}\rangle\vert^2 &=\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\sum_{m\in\mathbb{Z}^2}\,\Bigl|\int_{\mathbb{R}^2}\hat{f}(\xi)\hat{\psi}({(M_{j,k}^{-1})}^T\xi)e^{-2\pi i\langle m,{(M_{j,k}^{-1})}^T\xi\rangle}\mathrm{d}\xi\Bigl|^2&&\\\nonumber &=\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\sum_{m\in\mathbb{Z}^2}\,\Bigl|\int_{M_{j,k}^T([-\frac{1}{2},\frac{1}{2}]^2)}\,\hat{f}(\xi)\hat{\psi}({(M_{j,k}^{-1})}^T\xi)e^{-2\pi i\langle m,{(M_{j,k}^{-1})}^T\xi\rangle}\mathrm{d}\xi\Bigl|^2&&\\\nonumber &=\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\text{det}({(M_{j,k}^{-1})}^T)\sum_{m\in\mathbb{Z}^2}\,\Bigl|\int_{[-\frac{1}{2},\frac{1}{2}]^2}\hat{f}(M_{j,k}^T\xi)\hat{\psi}(\xi)e^{-2\pi i\langle m,\xi\rangle}\mathrm{d}\xi\Bigl|^2. \end{align} Since $\{e^{-2\pi i\langle m,(\cdot)\rangle}\,\vert\,m\in\mathbb{Z}^2\}$ in an orthonormal basis for $L^2([-\frac{1}{2},\frac{1}{2}]^2)$, we can apply the Parseval equality. Hence \begin{align} \nonumber \sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\sum_{m\in\mathbb{Z}^2}\,|\langle \hat{f},\hat{\psi}_{j,k,m}\rangle\vert^2&=\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\text{det}({(M_{j,k}^{-1})}^T)\int_{[-\frac{1}{2},\frac{1}{2}]^2}\vert\hat{f}(M_{j,k}^T\xi)\vert^2\vert\hat{\psi}(\xi)\vert^2\mathrm{d}\xi&&\\\nonumber &=\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\int_{M_{j,k}^T([-\frac{1}{2},\frac{1}{2}]^2)}\vert\hat{f}(\xi)\vert^2\vert\hat{\psi}({(M_{j,k}^{-1})}^T\xi)\vert^2\mathrm{d}\xi&&\\\nonumber &=\int_{M_{j,k}^T([-\frac{1}{2},\frac{1}{2}]^2)}\vert\hat{f}(\xi)\vert^2\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\vert\hat{\psi}({(M_{j,k}^{-1})}^T\xi)\vert^2\mathrm{d}\xi&&\\\nonumber &=\int_{\mathbb{R}^2}\vert\hat{f}(\xi)\vert^2\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\vert\hat{\psi}({(M_{j,k}^{-1})}^T\xi)\vert^2\mathrm{d}\xi. \end{align} We now show, that $\sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\vert\hat{\psi}({(M_{j,k}^{-1})}^T\xi)\vert^2=1$, which almost completes the proof. By the definition of $M$ and the classical shearlet we have \begin{align} \nonumber \sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\vert\hat{\psi}({(M_{j,k}^{-1})}^T\xi)\vert^2&= \sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\vert\hat{\psi_1}(2^{-j}\xi_1)\vert^2\vert\hat{\psi_2}(2^{\frac{1}{2}j}\frac{\xi_2}{\xi_1}-k)\vert^2&&\\\nonumber &=\sum_{j\in\mathbb{Z}}\vert\hat{\psi_1}(2^{-j}\xi_1)\vert^2\sum_{k\in\mathbb{Z}}\vert\hat{\psi_2}(2^{\frac{1}{2}j}\frac{\xi_2}{\xi_1}+k)\vert^2=1. \end{align} Finally we conclude again by Plancherel \begin{align} \nonumber \sum_{j\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}\sum_{m\in\mathbb{Z}^2}\,\vert\langle f,\psi_{j,k,m}\rangle\vert^2=\int_{\mathbb{R}^2}\vert\hat{f}(\xi)\vert^2\mathrm{d}\xi=\vert\vert\hat{f}\vert\vert^2=\vert\vert f\vert\vert^2. \end{align}