Subspace representation of registration and reconstruction in multi-frame Super-Resolution

Multi-frame Super-Resolution Reconstruction (SRR) problem has two bottlenecks; under constrained reconstruction and observation registration. Currently there is no practical solution for these problems, even in domain-specific cases. In this work we suggest solutions for both of these problems. SRR problem is transformed to a reduced space to render an over-constraint problem. Thus significant amount of computational saving and robustness is obtained. Moreover instead of estimating registration parameters we follow a more direct way and employ deformable models to register observations to the reference frame. By combining these two solutions we obtain faster, more flexible and superior SRR results.