CRM: Centro De Giorgi

This is the old version of the CRM site. Please use the new site on the page crmdegiorgi.sns.it

logo sns
Mathematical and Computational Aspects of Machine Learning

Unrolled ISTA and convolutional neural networks for limited-angle tomography reconstruction

speaker: Mathilde Galinier (Università degli studi di Modena e Reggio Emilia)

abstract: Computed Tomography makes use of computer-processed combinations of many X-ray measurements of an object, taken from different angles, and attempts to recover the inner structure of the object from the data. In the case of limited-angle tomography, the reconstruction problem is severely ill-posed and the traditional reconstruction methods, e.g. filtered backprojection (FBP), do not perform well. In this work, we investigate a brand-new method for limited-angle tomography reconstruction, based on the unrolled version of the ISTA algorithm where each iteration contains a convolutional neural network (CNN). The idea of this project has emerged from the observation that the backprojection operator can be approximated by a sequence of convolutions applied to the original object in the wavelet domain. Thus, each CNN has the same structure as the previously mentioned sequence of convolutions.


timetable:
Wed 9 Oct, 17:20 - 17:40, Aula Dini
<< Go back