Data-driven thresholding in denoising with Spectral Graph Wavelet Transform

This paper is devoted to adaptive signal denoising in the context of Graph Signal Processing (GSP) using Spectral Wavelet Transform (SGWT). issue addressed via a data-driven thresholding process transformed domain by optimizing parameters sense Mean Square Error (MSE) Stein’s Unbiased Risk Estimator (SURE). The SGWT considered built upon partition unity making transform semi-orthogonal so that optimization can be performed domain. However, since over-complete, divergence term SURE needs computed correlated noise. Two strategies called coordinatewise and block are investigated. For each them, derived for whole family elementary functions among which soft threshold James–Stein threshold. multi-scales analysis shows better performance than most recent methods from literature. That illustrated numerically series signals on different graphs.