We propose a variational approach to obtain super-resolution images frommultiple low-resolution frames extracted from video clips. First thedisplacement between the low-resolution frames and the reference frame arecomputed by an optical flow algorithm. Then a low-rank model is used toconstruct the reference frame in high-resolution by incorporating theinformation of the low-resolution frames. The model has two terms: a 2-normdata fidelity term and a nuclear-norm regularization term. Alternatingdirection method of multipliers is used to solve the model. Comparison of ourmethods with other models on synthetic and real video clips show that ourresulting images are more accurate with less artifacts. It also provides muchfiner and discernable details.
展开▼
