Signal denoising using line-adaptive lifting wavelet transform

Abstract

This paper describes denoising methods of 1D signals using soft thresholding of wavelet transform coefficients. Some adaptive techniques based on classical and lifting versions of the wavelet transform are implemented in software. A new method based on the line-adaptive "update first" lifting scheme is implemented and compared with scale-adaptive denoising based on classical and lifting wavelet transforms. The results show that the line-adaptive "update first" algorithm gives the best results. However, the least-squares (SNR) efficiency of all the methods is very similar. In non-adaptive techniques the denoising quality strongly depends on the proper choice of decomposition filter length according to the signal characteristics. Therefore, application of the adaptive schemes is signal independent. Computer experiments reveal that the line-adaptive "update-first" lifting signal denoising is characterised by very good SNR and the lowest maximum-absolute reconstruction error.