Image Denoising Based on Wavelet Domain Spatial Context Modeling
Xuchao Li, Shan'an Zhu · 2006
Using prior knowledge about the spatial clustering of the wavelet coefficients, a new image denoising method that applies the Bayesian framework is proposed. Wavelet coefficients of image are characterized by a two-state Gaussian mixture model (GMM), while their local spatial interactions are modeled by a Markov random field (MRF) model. The Expectation Maximization (EM) algorithm is used to estimate the parameters of the GMM, and an iterative updating technique known as iterative conditional modes (ICM) is applied to optimize the binary labels containing the positions of those wavelet coefficients that represent the useful signal in each subband. For each wavelet coefficient a shrinkage factor is finally determined, depending on its initial shrinkage factor and on the local spatial neighborhood in the label field. The qualitative and quantitative experimental results show that the new scheme outperforms other wavelet denosing methods, such as yielding significantly superior image quality, increasing peak signal-to-noise ratio (PSNR)