Blind signal separation by combining two ICA algorithms: HOS-based EFICA and time structure-based WASOBI

The aim of this paper is to combine the strengths of two recently proposed Blind Source Separation (BSS) algorithms. The first algorithm, abbreviated as EFIGA, is a sophisticated variant of the well-known Independent Gomponent Analysis (IGA) algorithm, FastlGA. EFIGA is based on minimizing the statistical dependencies between the instantaneous (marginal) distributions of the estimated source signals and therefore disregards any possible time structure of the sources. The second algorithm, WAS OBI, is a weight-adjusted variant of S OBI, a popular BSS algorithm that uses only the time structure of the source signals to achieve the separation. The separation accuracy of EFIGA and WASOBI can be assessed using the estimated source signals alone, therefore allowing us to choose the most appropriate of the two in every scenario. Here, two different EFIGA-WASOBI combination approaches are proposed and their performance assessed using images and simulated signals.