Modeling the autocorrelation of wavelet coefficients for image denoising
Huipin Zhang, Aria Nosratinia, Raymond O. Wells · 2002
The undecimated discrete wavelet transform (UDWT) is a powerful image denoising tool, well-known to perform better than orthogonal wavelets. Unlike the case of orthogonal wavelets, noise as well as signal in the UDWT domain are non-white. Because of this inter-pixel correlation, scalar operations such as thresholding do not take full advantage of the power of UDWT. In this paper, we present a model for autocorrelations in the UDWT domain, and use it in a Wiener-type denoising algorithm. This algorithm accounts for colored signal and noise, and also aims to implicitly match the directional information due to the local edges in the image.