We have presented a recipe for approximating the MSE associated with an estimate of an object based on blurred, Poisson data. We give an explicit formula , Eq. (12), for the bound when the bias in the estimate is known, as is the case with a Wiener filter estimate.
When the bias for an estimate, b, is unknown we can only suggest an approach. The suggestion is to find a Wiener filter estimate which yields the same residual noise power in data space as b, averaged over a local region. The composite Wiener-based bound may be a reasonable approximation to the exact calculation which requires knowledge of the bias. We intend to pursue this suggestion in future work; and to pursue a direct calculation of the analytic form of the bias for the RL algorithm, based on the residual error in data space.