Abstract:

A complete adaptive algorithm of denoising by wavelet transforms was developed, which has been presented by Xu,et al.The primary idea is that the wavelet transforms of signals are deeply correlated at several adjacent scales, while the noise is not. The form of noise standard deviation at each scale was given here,and then a method is presented to select the power threshold of noise. An effective standard deviation estimation of noise involved in a true signal was also described here, which was based on the fact that in fine scales wavelet transforms are dominated by noise. The two finest scales are correlated to estimate noise covariance better. The simulation shows that while getting rid of noise mostly, this algorithm keeps the main edges of signal well.

Key words: filtering, wavelet transform, threshold, spatial related