Performance study of the SURE trimmed denoising method for one dimensional signals
Maha A. Hassanein, Magdy Tawfik Hanna · 2017
An adaptive data driven threshold is proposed for denoising one dimensional signals. The threshold is derived in a SURE (Stein Unbiased Risk Estimator) based framework using trimmed thresholding and the wavelet coefficients are obtained by a Translation Invariant transform. A detailed mathematical derivation of a hybrid scheme of the Universal threshold and the SURE threshold with trimmed thresholding is provided. An optimized selection of the trimmed thresholding parameter (alpha) at each decomposition level of the transform is tackled and its effect on the overall performance of the proposed method is investigated. The experimental results show that the proposed method with automatic parameter adjustment is typically better or at least not worse than other wavelet based and non-wavelet based methods. It has been found that the method with optimal parameter value outperforms that with fixed parameter value more than 75% of the time over a wide range of input SNRs.