Brain Source Localization using Constrained Low Rank Canonical Polyadic Decomposition

A new tensor-based source localization algorithm is presented in this paper. It is a single-step algorithm in which tensor decomposition with an efficient rank estimation and source localization are performed in only one single step. Contrary to the previous single-step tensor-based STS-SISSY (Space-Time-Spike Source Imaging based on Structured Sparsity) method recently proposed by our group, the proposed method is robust to tensor over-factoring and gives more accurate results. In addition to the structural constraints on the sources required for their localization, group sparsity constraints on the loading over-estimated matrices of the constructed STS tensor is used to estimate its rank. The numerical results show the efficiency of the proposed method over the STS-SISSY one.