Abstract
Restoring an observed image suffering from blur and noise simultaneously is a challenging problem that may cause a large estimation error of blur and noise parameters. In this paper, a novel blind image deconvolution approach based on noise variance estimation is presented. This method first performs noise variance estimation from a noisy blurred image. Then, using the property that a certain type of blur may lead to a specific frequency component distortion of the image Fourier spectrum, the blur type can be reorganized. After this, according to the reorganized blur type, the image and blur model coefficients can be computed more efficiently by minimizing an objective function based on the autoregressive moving average (ARMA) model, where the maximum-likelihood (ML) method is used. The restored image is obtained with a least-squares filter. Experiments on images are presented, which show that the proposed method is capable of yielding good restoration results.
