Multi-frame super-resolution reconstruction is a technology which obtains a high-resolution image from several low-resolution images of the same scene. Among various super resolution methods, the regularized method is widely used since it has advantages for solving the ill-posed problems. However, the super-resolution reconstruction results based on this method strongly depend on the estimation accuracy of the optimum estimator. In this paper, a double-threshold Huber norm based maximum likehood estimator is proposed, which improves the threshold tolerance of the estimator and increases the estimation accuracy. Then a regularized algorithm based on this estimator is presented. The super-resolution reconstruction results of synthetic low resolution images confirm that the proposed algorithm has better performance over the existing algorithms. The proposed algorithm is also used to deal with the low-resolution images obtained from a plenoptic camera. The results confirm the effectiveness of the proposed algorithm.