Wavelet-domain Image Restoration Model Based on Variation Regularization
Caixia Li, Chanjuan Liu, Hailin Zou · International Journal of Signal Processing Image Processing and Pattern Recognition · 2014
There are two disadvantages in variational regularization based image restoration model.Firstly, the restored image is susceptible to noise because the diffusion coefficient depends on image gradient.Secondly, in the process of energy minimization, the selection of Lagrange multiplier which is used to balance the regular term and the fidelity term can directly affects the quality of the restored image.To solve the above problems, the multi-resolution feature of multi-scale wavelet is introduced into the energy minimization model and a wavelet based image restoration model is proposed.In the proposed model, Lagrange multiplier is replaced by an adaptive weighting function j which is constructed by the image wavelet transform coefficients.Theoretical analysis and experiment results show that, comparing with other methods, the proposed model reduces iterations in the energy minimization process, overcomes the cartoon effects in variational model and pseudo-Gibbs effect in traditional wavelet threshold methods, and can well protect the detail features while denoising.