Impact of sensor head geometry on the performance of hard-field tomography reconstruction from limited views

The sinusoidal Hough transform (SHT) approach for tomography sensors allows adequate quality of imaging from a severely limited number of measurements, typical for cases of limited access. We have studied of the performance of SHT reconstruction by 3 different error metrics in the case of sensor heads with a decreasing number of path integral measurements - from 52 to 28, which is essential for hard-field tomography applications in industry. The grouping of these measurements into parallel beam angular projections and robustness to noise is studied for several sensor head geometries, comprising of only 3 or 4 angular views. A SHT-based algorithm is applied on 3 objects with variations in their definition (diffuse vs abrupt boundaries), as well as topology. The reported results indicate that with a fixed small number of measurements the importance of the number of angular projections overrides the density of sampling within each angular projection.