Practical error bounds for binary tomography
Faculty of Sciences. Physics
1st International Conference on Tomography of Materials and Structures (ICTMS), Ghent, Belgium
University of Antwerp
The purpose of computed tomography (CT) is to compu te an accurate approximation of a scanned object from a series of its projections. As the ground tru th is typically unknown, it is not straightforward to determine the quality of such an approximation. Eve n if the reconstructed image corresponds almost perfectly with the observed projection data, it may still be quite different from the original object if the number of projections is small. We have recently developed a series of mathematical error bounds that provide quantitative guarantees on the quality of the reconstruction of a homogeneo us object (i.e. a binary image). As these bounds ar e based on idealized assumptions of the imaging model (assuming perfect, noiseless data), they have to b e adjusted to be useful in practice. In this article we show how one of these error boun ds can be adapted to be useful for bounding the qua lity of experimental images. Our experimental results su ggest that even though approximations have to be made due to noise and other errors in the data, the resulting bounds can still provide guidance on estimating the reconstruction quality.