An improved constrained ICA with reference based unmixing matrix initialization

The constrained independent component analysis (cICA) provides a general framework to incorporate the prior information. To overcome the unstable problem encountered in the cICA algorithm of [2], an improved method is proposed in this letter by using the reference based unmixing matrix initialization. Compared to the original algorithm, the prior information of references is not only used in the contrast function, but also used in the initialization. The utility of the proposed method is demonstrated by the experimental results on the artificial data and the sound signals.