This paper describes an iterative image restoration technique that closely resembles the R-L iteration but that avoids the amplification of noise that occurs in the R-L iteration. I call this the ``damped'' R-L iteration because the modification appears as a damping factor that slows changes in regions of the model image that fit the data well while allowing the model to continue to improve in regions where it fits the data less well.
The approach used is to modify the likelihood function so that it is ``flatter'' in the vicinity of a good fit. It is unlikely that this ad hoc approach represents the best solution to the noise amplification problem, but it does have the advantages that it is easy to implement and robust, and it appears to produce better results than the R-L method in many cases. The method produces restored images that have good photometric linearity and little bias.