1. The document proposes a robust super-resolution algorithm that minimizes a Gaussian-weighted L2 error norm. This suppresses the influence of intensity outliers without requiring additional regularization.
2. The algorithm is based on maximum likelihood estimation but uses a Gaussian error norm instead of a quadratic norm. This makes the algorithm robust against outliers by reducing their influence to zero.
3. The effectiveness of the proposed algorithm is demonstrated on real infrared image sequences with severe aliasing and intensity outliers, where it outperforms other methods in handling outliers and noise.