Robust super-resolution reconstruction based on adaptive regularization

A robust super-resolution reconstruction algorithm with Huber norm and bilateral total variation is proposed in this paper. The Huber norm is adopted for data fidelity term instead of L1 or L2 norm to improve robustness to outliers. And also an adaptive method updating the regularization parameter simultaneously with the restored image is proposed. Compared with most of the present approaches selecting the parameter manually, the adaptive algorithm can improve performance and avoid the randomness and subjectivity of experiments. Simulation results of both synthetic data and real data confirm the robustness of the proposed algorithm and its superiority to other algorithms.