A new approach to the reconstruction of images from Radon projections (Q859623)

A new algorithm for image reconstruction in the closed unit disk \(\mathbb{D}=\{(r,\theta): 0\leq r\leq 1, \, 0\leq \theta\leq 2\pi\}\) based on Radon projections is provided. One begins with the Chebyshev polynomials \(U_k\) defined for \(x=\cos\varphi\) by \(U_k(x)=\sin(k+1)\varphi/\sin\varphi\). They are orthogonal with respect to the weight \(\sqrt{1-x^2}\) on \([-1,1]\). The polynomials \(U_{j,k}(x,y)=U_k(\theta_{j,k};x,y)\equiv U_k(x\cos\bigl(\frac{j\pi}{k+1}\bigr)+y\sin\bigl(\frac{j\pi}{k+1}\bigr))\), \(0\leq j\leq k\) and \(0\leq k\leq n\) are shown to form an orthonormal basis for the polynomials of degree at most \(n\) with respect to the \(L^2(\mathbb{D})\) inner product. The orthogonal projection of \(f\in L^2(\mathbb{D})\) onto such polynomials of degree \(2m\) can be expressed as \[ S_{2m}(f;x,y) =\sum_{\nu=0}^{2m} \frac{1}{\pi} \int_{-1}^1 {\mathcal{R}}_{\varphi_\nu} (f;t) \Phi_\nu(t;x,y)\, dt \] in which \({\mathcal{R}}_\varphi\) denotes the line integral of \(f\) in the \(\varphi\)-direction, \(\varphi_\nu = 2\pi\nu/(2m+1)\), and \[ \Phi_\nu(t;x,y) = \frac{1}{2m+1} \sum_{k=0}^{2m} (k+1) U_k(t) U_k(\varphi_\nu;x,y). \] Using appropriate quadrature formulas, one can then define a projection \(\mathcal{A}_{2m}(f)\) that only uses the Radon data \(\mathcal{R}(f;\cos(j\pi/(2m+1)))\) and still reproduces polynomials of degree at most \(2m-1\). The approximation algorithm amounts to computing \(\mathcal{A}_{2m}(f)\) from the appropriate Radon data via quadrature. The rigorous estimate \(\| \mathcal{A}_{2m}\| _\infty\approx m\log(m+1)\) is provided, where \(\| \cdot\| _\infty\) denotes the operator norm on \(C^2(\mathbb{D})\). In particular, \(\mathcal{A}_{2m}(f)\) converges uniformly to \(f\) if \(f\in C^2(B)\).