Abstract
We present an iterative method for reducing the blurring and roughness artifacts in cone-beam CT reconstructions that are caused by unknown geometric jitter. Geometric uncertainties are estimated by registering directly and indirectly filtered projections. We assume that the jitter effects can be largely compensated for by a proper software translation of the detector within its plane. First results on simulated data indicate a noticable improvement of image quality where this assumption is fullfilled (backprojection mismatch (BPM) decreased by 82%), and also where it is not fulfilled (BPM decreased by 78%). The method is robust with respect to image noise and can also handle truncated data.