Bias Removal Techniques for Component Pursuit

Abstract

The component pursuit problem is introduced under blind environment when Gaussian noise is present. An improved quantitative measure of non-Gaussianity, called Gaussian moments, is deduced correspondingly. After analyzing the useful property of Gaussian moments, an objective function is presented which can be suitable in the noisy context. As a result, a one-unit algorithm is presented with bias removal for quasi-whitened data. Computer simulations illustrate the better performance of the proposed approach.