A penalised data-driven block shrinkage approach to empirical Bayes wavelet estimation

In this paper we propose a simple Bayesian block wavelet shrinkage method for estimating an unknown function in the presence of Gaussian noise. A data-driven procedure which can adaptively choose the block size and the shrinkage level at each resolution level is provided. The asymptotic property of the proposed method, BBN (Bayesian BlockNorm shrinkage), is investigated in the Besov sequence space. The numerical performance and comparisons with some of existing wavelet denoising methods show that the new method can achieve good performance but with the least computational time.