BAYESIAN RECONSTRUCTION BASED ON FLEXIBLE PRIOR MODELS

A new approach to Bayesian reconstruction is proposed that endows the prior probability distribution with an inherent geometrical flexibility, which is achieved through a transformation of the coordinate system of the prior distribution or model into that of the reconstruction. With this warping, prior morphological information regarding the object that is being reconstructed may be adapted to various degrees to match the available measurements. The extent of warping is readily controlled through the prior probability distributions that are specified for the warp parameters. The complete reconstruction consists of a warped version of the prior model plus an estimated deviation from the warped model. Examples of tomographic reconstructions demonstrate the power of this approach.