Optimization Of Compound Regularization Parameters Based On Stein'S Unbiased Risk Estimate

Recently, the type of compound regularizers has become a popular choice for signal reconstruction. The estimation quality is generally sensitive to the values of multiple regularization parameters. In this work, based on BDF algorithm, we develop a data-driven optimization scheme based on minimization of Stein's unbiased risk estimate (SURE) statistically equivalent to mean squared error (MSE). We propose a recursive evaluation of SURE to monitor the MSE during BDF iteration; the optimal values of the multiple parameters are then identified by the minimum SURE. Monte-Carlo simulation is applied to compute SURE for large-scale data. We exemplify the proposed method with image deconvolution. Numerical experiments show that the proposed method leads to highly accurate estimates of regularization parameters and nearly optimal restoration performance.