Feature Extraction Based on Nonparametric Waveform Estimation and Matching Pursuit

This paper presents a novel method for extracting feature waveforms of the compound signal, which combined the nonparametric waveform estimation based on filter bank with the conventional matching pursuit (MP). At each iteration of the MP, the proposed method expressed the extracted feature waveform that has the best matched with the signal local structure with a series expansion of a set of nonparametric basis functions which generated by filtering the template signal through a bandpass filter bank. A prior signal information is no more require for a template signal owe to apply the adaptive template signal in this method. So, the construction of general described by some parameters is unnecessary and it felicitates the method for a wide variety of applications. Simulated and experimental results verify the practicability and effectiveness of this algorithm, even if the noise frequency band coincides with that of the feature waveform of the signal.