No abstract

Computed Tomography is increasingly employed for nondestructive evaluation, with the aim of reconstructing a surface mesh of a scanned object from radiographic projections. State-of-the-art algorithms first reconstruct a voxel grid and then extract a surface mesh using existing meshing algorithms, often leading to stair-like aliasing artifacts along the grid axes, due to the grid's orientation-dependent resolution. We circumvent such artifacts in filtered backprojection reconstructions by optimizing the mesh's vertex positions using information taken directly from the projections, rather than from a voxel grid. We show that our approach reduces stair artifacts both visibly and measurably, at relatively little additional computational cost. Our method can be tied into existing mesh extraction algorithms and removes stair artifacts almost entirely.

In industrial CT, objects are often reconstructed from a circular scan trajectory. This reconstruction is possible even if only half of the object is visible in the detector at any given time, but at the cost of redundancies that can be used for the correction of miscalibrations. We propose a scheme for recalibration using only a sinogram of projections truncated in such a manner. The sinogram does not need to be of any particular phantom and can depict any object. We show that the scheme is effective and produces calibrations that are similarly accurate as their untruncated counterparts.

In computed tomography (CT) reconstruction, ring artifacts emerge from incorrectly normalized or defective detector elements. Correction algorithms often introduce blur or do not correctly accommodate the behaviour of those artifacts. Normalization errors stem from noise in the detector images during the normalization process and are always present to some degree. We propose a method for correcting ring artifacts from incorrectly normalized detector elements in the sinogram. Our approach compensates for errors both in the individual gain as well as offset of pixel values. We reduce blur by inferring gain and offset information for each pixel from its neighbors only in a subset of all projections. We show with datasets from real measurments that our method is effective at mitigating the shortcomings of purely offset based approaches and approaches using all projections. Furthermore, our method can be efficiently implemented compensating for most overhead times. Under usual circumstances, our method can be implemented to function with no additional time overhead at all.