Recursive CR-Type Bounds and the EM Algorithm: Applications to ECT Image Reconstruction

We give a class of iterative algorithms to monotonically approximate submatrices of the CR matrix bound on the covariance of any estimator of a vector parameter/spl theta/. A natural implementation of the iterative algorithm employs a" complete data - incomplete data" formulation similar to that underlying the EM parameter estimation algorithm. Our results make it feasible to compute CR-type bounds for previously intractable problems involving a large number of "nuisance parameters," such as arise in image reconstruction.