An adaptive LMS technique for wavelet polynomial threshold denoising
Sushanth G. Sathyanarayana, David Akopian, Sos С. Agaian · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2011
Threshold operators are conventionally used in wavelet-based denoising applications. Different thresholding schemes have been suggested to achieve improved balance between mitigating various signal distortions and preserving signal details. In general, these state-of-the-art threshold operators are nonlinear shrinkage functions such as well known "soft" and "hard" thresholds and their hybrids. Recently a nonlinear polynomial threshold has been introduced which integrates several known approaches and can be optimized using a least squares technique. While significantly improving the performance - this approach is computationally intensive and is not flexible enough for band-adaptive processing. In this paper an adaptive least mean squared (LMS) optimization approach is proposed and studied which drastically reduces computational load and is convenient for band-adaptive denoising scenarios. The approach is successfully applied to 1D and 2D signals, and the results demonstrate improved performance in comparison with the conventional methods.