Regularized mre method for blind multichannel image deconvolution
W. Souidene, Azeddine Beghdadi, Karim Abed‐Meraim, Belkacem Mouhouche · 2005
In this paper we deal with blind image restoration in a multi-channel framework. We use the mutually referenced equalizers (MRE) in order to perform exact image restoration in the noiseless case and under some identifiability conditions. In presence of noise, this method fails to restore correctly the original image. So, we introduce a regularization term to improve the restored image quality. We introduce two regularization techniques: a norm minimization scheme and an eigen value truncation scheme in order to perform the regularization and reduce the noise amplification assigned to image restoration. We compare the performance gain of our method for different number of observed images with the mono-channel case. We evaluate the performance of our method using an image similarity index.