Noise estimation in signal restoration using regularization
Adam Thompson, Jim Kay, D. Michael Titterington · Biometrika · 1991
Several estimators of variance are compared in the context of problems where smoothing is incorporated in the estimation of regression functions. Many of the estimators are scaled quadratic forms in the data, and some incorporate steps that try to accommodate substantial nonconstancy in the response function. The estimators are described and an account is provided of an empirical study based on four one-dimensional source functions of varying degrees of irregularity.