Based on information theoretic approach and projection pursuit approach, Hyva??rinen and Oja have introduced a family of novel contrast functions for independent component analysis (ICA) and proposed a Fast-ICA algorithm. Fast-ICA algorithm converges fast but it is not robust for colored sources such as biomedical signals. By optimizing Fast-ICA's contrast function, and applying natural Riemannian gradient in Stiefel manifold, we present in this paper a modified algorithm which can deal with colored signals successfully. Further computer simulation results and comparisons demonstrate the effectiveness and validity of our algorithm.