Adaptive regularization for resolution restoration of multiframes considering subpixel contributions

In this work we propose an adaptive resolution restoration algorithm for sequence of images that considers the subpixel contribution of each frame. The Regularized Least-Squares (RLS) algorithm is modified to include an extra regularization. In many previous works, it is considered regularization to mitigate only the degradation due to noise. The proposed algorithm also mitigates the distortions caused by the subsampling process. The contribution from additional frames is exploited by estimating the subpixel displacements. The pixels amplitudes from other frames, displaced by subpixel distances, provide additional information to mitigate degradations due to undersampling, like the aliasing. The extra regularization adapts according to the frames contributions. In the motion estimation step only the reliable displacement vectors are chosen for the restoration process. The proposed model significantly improves the objective (SNR) and the subjective (visual) image quality.