Orthogonal wavelet frames and vector-valued wavelet transforms (Q2381649): Difference between revisions

A pair of Bessel sequences \(\{x_j\}\) and \(\{y_j\}\) in a separable Hilbert space \(H\) is said to be orthogonal if the composition operator \( \sum_j \langle \cdot,\;x_j\rangle \;y_j=0\). The direct sum sequence \(\{x_j^1\oplus\cdots\oplus x_j^N\}\) is a Parseval frame for the \(N\)-fold direct sum \(H\oplus H\oplus\dots H\) if and only if \(\{x_j^k\}_j\) is a Parseval frame for \(H\) (meaning \(\sum_j \langle \cdot,\;x_j^k \rangle \;x_j = I\) for each \(k=1,\dots,N\)), and the sequences \(\{x_j^k\}_j\) are orthogonal for different \(k\). Given functions \(m_0,\dots, m_r\in L^\infty [0,1)\) let \(M(\xi)\) denote the \(r\times 2\) matrix whose first column is the transpose of \((m_0(\xi),\dots,m_r(\xi))\) and whose second column is that of \((m_0(\xi+1/2),\dots, m_r(\xi+1/2))\) and let \(\widetilde{M}\) be the submatrix obtained by deleting the first row of \(M\). A refinable function (one whose Fourier transform satisfies \(\widehat{\phi}(2\xi)=m_0(\xi)\widehat{\phi}(\xi)\) for some periodic \(m\)) is said to satisfy the unitary extension principle if, whenever the matrix \(M(\xi)\) satisfies \(M^\ast(\xi)M(\xi)=I_2\) then the affine system (\(\psi_{j,k}=2^{j/2}\psi(2^j x-k)\)) generated by \(\{\phi,\psi_1,\dots,\psi_r\}\) with \(\widehat{\psi}_k(2\xi)=m_k(\xi)\widehat{\phi}(\xi)\), defines a Parseval wavelet frame for \(L^2(\mathbb{R})\). Here the authors prove the following about \(\phi\) satisfying the unitary extension principle. Suppose that \(\mathcal{M}=\{m_0,m_1,\dots,m_r\}\) and \(\mathcal{N}=\{m_0,n_1,\dots,n_r\}\) are such that \(M^\ast M(\xi)=I_2=N^\ast N(\xi)\) and \(\widetilde{M}^\ast(\xi)\widetilde{N}(\xi)=0\) for almost all \(\xi\in [0,1)\). Then the affine systems generated by \(\psi_k\) and \(\eta_k\) where \(\widehat{\psi}_k(2\xi)=m_k(\xi)\widehat{\psi}_k(\xi)\) and \(\widehat{\eta}_k(2\xi)=n_k(\xi)\widehat{\eta}_k(\xi)\) form orthogonal Parseval wavelet frames. A more general construction of pairwise orthogonal multiwavelet frames is also provided and applied in higher dimensions. The purpose of these results is to define true vector-valued discrete wavelet transforms in which wavelet expansions do not reduce to coordinatewise expansions. Discrete implementations and applications to wavelet decomposition of color images are outlined.