Filtern
Dokumenttyp
Schlagworte
- Reconstruction algorithm (16) (entfernen)
Organisationseinheit der BAM
Computed tomography reconstructions of projected data usually assume an ideal point-like
focal spot. However, in practice, the actual finite focus size leads to blurred projections,
unavoidably. The elements (voxels) of the reconstruction array are smeared differently as a
function of their individual position relative to source and detector. Neglecting the focal
smearing generates considerably blurred reconstructions. Previous attempts were based on
directly deconvolving blurred projections from a constant kernel, which yields better results
but still exhibits artefacts. Iterative algorithms can be adopted to take into account focal
smearing as a function of position. However, this requires detailed prior knowledge of
quantitative smear functions. The DIRECTT (Direct Iterative Reconstruction of Computed
Tomography Trajectories) algorithm is a promising candidate to meet these requirements. It
has been demonstrated elsewhere that DIRECTT copes with limited data sets such as
limited view and region-of- interest data by tracing single sinusoidal-like trajectories in
Radon space, which are selected from the set of all possible trajectories by appropriate
criteria. Currently, position-dependent smearing is used in the projection part of iterations.
At the example of model reconstructions we demonstrate the gain of spatial resolution by
iterative variable desmearing according to the DIRECTT algorithm in comparison to the
standard filtered back-projection.
Statistical analysis of tomographic reconstruction algorithms by morphological image characteristics
(2010)
We suggest a procedure for quantitative quality control of tomographic reconstruction algorithms. Our task-oriented evaluation focuses on the correct reproduction of phase boundary length and has thus a clear implication for morphological image analysis of tomographic data. Indirectly the method monitors accurate reproduction of a variety of locally defined critical image features within tomograms such as interface positions and microstructures, debonding, cracks and pores. Tomographic errors of such local nature are neglected if only global integral characteristics such as mean squared deviation are considered for the evaluation of an algorithm. The significance of differences in reconstruction quality between algorithms is assessed using a sample of independent random scenes to be reconstructed. These are generated by a Boolean model and thus exhibit a substantial stochastic variability with respect to image morphology. It is demonstrated that phase boundaries in standard reconstructions by filtered backprojection exhibit substantial errors. In the setting of our simulations, these could be significantly reduced by the use of the innovative reconstruction algorithm DIRECTT.