Choosing the regularization parameter in image restoration
D. Michael Titterington · Lecture notes-monograph series · 1991
Many procedures in statistical image restoration can be regarded as regularization techniques involving a scalar smoothing parameter.The paper collates several methods for choosing the smoothing parameter, including model-based maximum likelihood.Minimum risk, generalized cross-validation, choice based on fit to the data, and "equivalent degrees of freedom" choice.Some theoretical and empirical comparisons are summarized.