Abstract:
Tomographical algorithms can be separated into two classes: analytical “one-step” methods and iterative reconstruction algorithms. Analytical methods are fast, but require projection data of high quality and are impossible to adapt to use prior knowledge about the reconstructed object. Iterative algorithms have less strict requirements for the used data and are more flexible, but their computation time impedes real-time reconstructions. By reformulating the reconstruction problem as a classification problem, a third option becomes available: machine learning. Feedforward neural networks are especially interesting because of the relative short computation time required to calculate output values. By processing examples, neural nets can be trained to perform reconstructions. This solution offers reconstruction quality comparable to iterative algorithms and computation time in the same order of magnitude as analytical methods.
Publications:
E. Janssens, De Beenhouwer, J., Van Dael, M., Verboven, P., Nicolai, B., and Sijbers, J., “Neural Network Based X-Ray Tomography for Fast Inspection of Apples on a Conveyor Belt”, in IEEE International Conference on Image Processing, 2015, pp. 917-921.