基于离散小波变换的Overcomplete ICA并行结构

A Parallel Architecture Using Discrete Wavelet Transform for Overcomplete ICA

本文利用离散小渡变换提出了一种过完备独立成分分析(Overcomplete ICA)的禀性结构.它是一个由两个子 Overcomplete ICA 过程组成的混合系统。其中一个过程将高频的小波部分作为输入,另一个过程将低频的部分作为输入。这两个过程的输出结果最后被合并为最终的结果。对比现有的 Overcomplete ICA 算法,本文提出的方法利用了全部的观测信息.而两个子过程的有效输入长度仅为原来的一半。因此,本文提出了一种处理 Overeomplete ICA问题的新途径。文中的实验数据显示,通过此方法可以成功地分离混合的声音数据。

This paper utilizes a discrete wavelet transform to present a parallel architecture for overcomplete independent component analysis （Overcomplete ICA）, which is a hybrid system for consisting of two sub-overcomplete ICA processes. One process takes the high frequency wavelet part of observations as its inputs, meanwhile the other process takes the low-frequency part. Their results are then merged to generate the final results. Compared to the existing overcomplete ICA algorithms, the proposed approach utilizes the full observation information, but the effective input length of the two parallel processed is halved. Therefore it generally provides a new way for overcomplete ICA implementation. In this paper, experimental result has shown its success in extracting the in separating the mixed speech signals.