DOI: 10.5176/2251-1911_CMCGS14.17
Authors: Zenith Purisha and Samuli Siltanen
Abstract: A new approach in tomographic inversion using
Non-Uniform Rational B-Splines (NURBS) combined with
Markov Chain Monte Carlo is discussed. Low dimension of
parameters is the benefit in using NURBS, but the resulting
inverse problem is nonlinear. Markov Chain Monte Carlo comes
forth to tackle this problem. Another advantage is that the
result will be directly in CAD software so that it will be
convenient for optimizing the shape. Numerical example with
simple simulated data, a simple homogeneous simple shape with
attenuation one inside the curve and zero outside the curve is
given. The result is compared with filtered back projection and
Tikhonov regularization. The potential drawback of the proposed
method is heavy computation.
Keywords: tomographic; NURBS; Bayesian inversion;
MCMC
Updating...