Birmingham-Southern College, Computer Science, Birmingham, AL 35254
Density image reconstruction empowers us to see the unseen, whether the subject is brain tissue or rock strata. However, its costs are also great: radiation exposure for a patient; time and resources for a geological crew. We sought to develop efficient and accurate algorithms for collecting measurement data and reconstructing a given two-dimensional image. In this interest, we modeled a measurement tool firing along a straight line (beam) and tested it in different arrangements on a virtual cross-section density image. In addition, we implemented two different solver methods, gradient descent and matrix decomposition, to compare reconstructions on the collected measurement values. Algorithms for reconstruction were compared on accuracy to the original image, number of measurements taken, and overall computation time. Results showed that gradient descent was far superior to matrix decomposition in all categories as a solver. As for beam algorithms, parallel projections fired at varying theta angles proved to be the most effective single algorithm. However, our best results were a combination of parallel projections followed by a center-point projection that targeted blurred interior areas.