Nonstationary blur modeling using robust eigenkernels

In this paper we propose a robust low rank model for restoring images corrupted by a nonstationary blur. This work improves the eigenkernels model proposed by Gwak and Yang. Their method uses standard Principal Component Analysis (PCA), and is thus not well suited to data with outliers. We replace PCA by its robust version. Numerical experiments show that our proposal offers reliable blur description and restoration even in the presence of salt-and-pepper noise, in which the original framework fails.