Un algorithme primal-dual stochastique et ses applications à la reconstruction d’images pour la tomographie à émission de positrons
Claire Delplancke
University of Bath
https://clairedelplancke.github.io/
Date(s) : 03/06/2022 iCal
14h30 - 15h30
L’algorithm
A stochastic primal-dual algorithm and its applications to image reconstruction for positron emission tomography
The SPDHG (Stochastic Primal-Dual Hybrid Gradient) algorithm is a stochastic version of the PDHG (Primal-Dual Hybrid Gradient) algorithm developed by Chambolle and Pock, which is used in inverse problems where the data attachment term and the regularizer are convex but not necessarily smooth. Thanks to its randomized component, SPDHG allows to perform only partial evaluations of the direct operator and its adjoint. This makes it a particularly suitable algorithm for positron emission tomography (PET), where the main obstacle to the practical adoption of sophisticated iterative methods is the computational cost of projections. I will present a convergence result for SPDHG as well as applications, in particular related to the issue of the choice of the step, on real and simulated PET data sets.
Emplacement
I2M Chateau-Gombert - CMI, Salle de Séminaire R164 (1er étage)
Catégories