Using Sub-Band Reconstruction in Wavelet Space and Fourier Transform to Extract Local Features from Analytical Signals Exactly and Straightforwardly

Abstract

To cope with the problem of frequency aliasing in Mallat algorithm, which makes traditional discrete wavelet transform (DWT) inappropriate for feature extraction in some cases, an improved algorithm composed of sub-band reconstruction and Fourier transform is suggested through which the original signal could be split into a series of sub-bands of different frequencies with little distortion both in time and frequency domains. This strategy is developed to extract local features from analytical signals accurately as well as straightforwardly. Some NIR spectra have been selected as examples to demonstrate the availability and application of the proposed method.

Keywords:

• Discrete wavelet transform (DWT)
• Feature extraction
• Fourier transform
• Frequency aliasing
• Mallat algorithm
• Near-infrared spectroscopy (NIR)
• Sub-band reconstruction

This work was supported by “973” National Key Basic Research Program (2007CB310500) and NSF (20775023, 20675028) of China.