Improved Denoising with Robust Fitting in the Wavelet Transform Domain

摘要

In this paper we present a new method for thresholding the coefficients in the wavelet transform domain based on the robust local polynomial regression technique. It is proven that the robust locally-weighted smoother excellently removes the outliers or extreme values by performing iterative reweighting. The proposed method combines the main advantages of multiresolution analysis and robust fitting. Simulation results show efficient denoising at low resolution levels. Besides, it provides simultaneously high density impulse noise removal in contrast to other adaptive shrinkage procedures. Performance has been determined by using quantitative measures, such as signal to noise ratio and root mean square error.