Blind source recovery: algorithm comparison and fusion
暂无分享,去创建一个
We study different solutions of the so‐called blind source recovery (BSR) problem: Given an m‐dimensional mixture random process x(t) and an m × n mixing matrix A, solve the underdetermined equation x(t) = As(t) + e(t) with the unknown n‐dimensional source process s(t) and noise e(t). For simplicity, we will assume that the processes are i.i.d. Moreover, s(t) has to satisfy additional assumptions such as independence or sparsity, and depending on these properties the above equation has a unique solution or reduced indeterminacies. We recall two BSR algorithms based on one of these two criteria. After discussing their properties, we show that they can be fused together in order to recover sources from a combined model. Simulations on artificial and real‐world data shows the feasibility of the proposed algorithm.
[1] Terrence J. Sejnowski,et al. Learning Nonlinear Overcomplete Representations for Efficient Coding , 1997, NIPS.
[2] Fabian J. Theis,et al. Robust overcomplete matrix recovery for sparse sources using a generalized Hough transform , 2004, ESANN.
[3] Fabian J. Theis,et al. Blind source separation and sparse component analysis of overcomplete mixtures , 2004, 2004 IEEE International Conference on Acoustics, Speech, and Signal Processing.
[4] Fabian J Theis,et al. Formalization of the Two-Step Approach to Overcomplete BSS , 2002 .