A practical stopping rule for iterative signal restoration

Abstract

The randomized generalized cross-validation (RGCV) criterion is developed as an efficient stopping rule for iterative signal restoration, and it is shown that it can be used on relatively small data sets with a large degree of confidence. Various experiments demonstrate the ability of the RGCV stopping rule to perform well for a large range of blur size ratios and noise levels, and for smaller signal lengths. The RGCV stopping rule has several advantages over previous methods. It is computationally efficient and easy to implement, taking advantage of quantities already computed in the restoration algorithm. It does not require any knowledge of the noise variance. It provides a restoration closer to the ideal restoration than the residual method. >