A Higher Order Regularization Approach for Object Reconstruction with Mixed Laplace-Gaussian Likelihood

A combined first and second order variational model is proposed for reconstructing images corrupted by mixed Laplace-Gaussian noise. The model is constructed by joint maximum a posteriori estimation and expectation maximization. Numerical algorithm is studied by integrating splitting technique into augmented Lagrangian method with modification, such as introduction of adaptively selective functions for preserving details of original images. An adaptive soft-shrinking formulation is advanced for mixed noise removal, in which an alternating minimization algorithm is established. Numerical experiments show validation in tomography reconstruction and image restoration.