Estimating Columns of Under-determined Mixing Matrix by Information Index Removal and Perturbed Mean Shift Algorithm

We proposed an inner point removal based on information index and perturbed mean shift algorithm to estimate the number of columns in the mixing matrix for under- determined blind source separation (BSS) with the unknown number of sources. The information index is used for preliminary filtering all relevant points clustered along the basic independent components. Then, the perturbed mean shift algorithm with entropy measure is applied to enhance to actual basic independent components. The experimental results on synthetic and real speech signals confirm the performance of the proposed method is more accurate than a standard k-mean method and AICA method.