Improving tomographic images with a filtered projection in real space

A certain quantity of noise, which reduces the diagnostic power of the tomographic technique, is always present in a reconstructed image. This is due to various reasons. On the one side, the finite precision of computers and the acquisition system geometry generate artefacts that degrade the image quality. On the other side, there are errors due to the experimental acquisition, the mathematical modelling of the physical problem and, last but not least, those caused by the choice of the reconstruction algorithm. In the present work this last aspect is improved using a reconstruction method known as filtered backprojection (FBP) in which the projections are filtered in real space rather than in Fourier space. In this way some of the so called aliasing artefacts are reduced, in particular those due to interperiodical interference of the F-transformed tails which represent one of the main problems while working in transformed space. The algorithm proposed in this paper represents an alternative to the usual filtering technique which consists of applying a filter, typically the Hamming filter, which manipulates the projections in transformed space in order to cut the high spatial frequency components. This is an important source of noise. The main limitation of such a procedure is loss of correspondence with real space and, consequently, has the possibility of introducing non-physical effects in the reconstructed image. © 2001 Elsevier Science Ltd. All rights reserved.