Wavelet-Based Image Denoising Using an Adaptive Anisotropic Markov Random Field Prior Model

摘要

This paper presents a new Markov Random Field prior model for configurations of wavelet coefficients in order to improve the performance of wavelet-based image denoising methods. The wavelet threshold algorithm is popular for the reduction of noise in images, which replaces wavelet coefficients with small magnitude by zero and keeps or shrinks the other coefficients. This procedure only makes use of the local effect of wavelet transform. Although a wavelet transform has decorrelated properties, the large coefficients corresponding to important image features are correlated. We therefore take into account dependencies between wavelet coefficients and propose a new prior model for configurations of them in a geometrical Bayesian framework. This model is an anisotropic model, which is adaptive to the wavelet subbands corresponding to three orientations in the image. By this prior model, the probability of being significant for each coefficient is computed and each modified coefficient is decided separately. The quantitative and qualitative results for test images demonstrate the improved denoising performance over related earlier methods.