Shift invariant wavelet denoising using interscale dependency

Abstract

Statistical modeling in the wavelet domain has proven its usefulness, as can be seen in image denoising. The index of PSNR and the visual quality are both improved, compared with Donoho and Johnstone’s wavelet threshold method. In [3], a pixel-adaptive Bayesian (PAB) denoising approach in the wavelet domain was proposed, which favorably compares with hidden Markov model (HMM) based approaches. However, the denoised images also suffer from the Gibbs-like artifacts, like ringing around the edges and speckles in the smooth regions. In this paper, we extend the PAB approach [3] to shift-invariant (SI) wavelet denoising in order to reduce these unpleasant artifacts. The experimental result shows that not only is the visual quality greatly improved but also a PSNR gain of about 0.7~0.9 dB is obtained. The proposed approach, called siPAB, outperforms siHMT, which is a competitive SI wavelet denoising approach, by 0.1~0.5 dB