KAMICA, a new deflation ICA algorithm based on Kurtosis alternating maximization.

In this article, a new Kurtosis based deflation Independent Component Analysis (ICA) algorithm is proposed. Contrary to conventional Kurtosis based deflation ICA algorithms where the separator vector is obtained by optimizing a given Kurtosis-based objective function using gradient-like approaches, the proposed algorithm performs a component-wise optimization of this separator vector through an alternating maximization strategy. This leads to an exact computation of each component of the separator vector by simply rooting a 4-th order polynomial derived from the objective function. The good performance of the proposed algorithm is confirmed and compared to two well known ICA algorithms, namely FastICA and RobustICA.