Video super-resolution based on block motion estimation and gradient magnitude

Abstract

Super-resolution (SR) technology aims to improve image resolution from a set of observed low resolution (LR) frames. This paper presents a new SR approach based on block motion estimation (ME) with gradient magnitude and the block types. ME is applied between a reference and its neighboring frames. If the motion is estimated inaccurately undesirable results are occurred at high resolution (HR) output frame. So, adaptive and static thresholds are used to eliminate improper blocks in registration. Outlier pixel detection is applied in reconstruction step. Finally, a deblurring process is performed on the HR frame. Proposed method results are compared with different interpolation methods. The experimental results are shown both peak-signal-to-noise ratio (PSNR) and as visually.