Moving objects is one of the main bottlenecks in many computer vision problems including video super resolution. Video Super-resolution algorithms reconstruct a high resolution video from a low resolution video. Recent advances in Super-Resolution techniques show trends towards model-based or example-based approaches. In this article a new method for video super-resolution problem based on our recent paper is proposed with at the same time handles moving objects efficiently. The overall method is based on this hypothesis, which some high-resolution(HR) images from the scene are available. Instead of the previous example based methods, which do blocking, here the whole training image is mapped to each frame. Proper mapping of the HR image onto LR frames has been done using a combination of feature-based and area-based registration methods. After creating a suitable mask for handling moving objects, each LR frame is fused with the registered HR image. The simulation results show the better performance of the proposed algorithm in compare with some other superresolution methods

Keywords