An Improved Method for the FastICA Algorithm
Feng Zhao, Min Cai, Yunjie Zhang · 2010
The FastICA algorithm based on Newton's iteration method can rapidly find hidden independent component from the mixed observations, and is widely used in the field of blind source separation. However, we need further improve the algorithm performance when processing massive data (such as image data). In this paper, an improved FastICA algorithm is proposed for blind source separation by establishing a Newton's iteration method with fifth-order convergence. The simulations show that, in contrast with FastICA algorithm, proposed algorithm has comparable separation performance and fewer iteration numbers.