Multi-frame example-based super-resolution using locally directional self-similarity

This paper presents a multi-frame example-based super-resolution (SR) algorithm using locally directional self-similarity. Existing example-based super-resolution algorithms generate patches using multiple training images or single self-image. On the other hand the proposed method minimizes the patch mismatching error by generating a patch dictionary in a local region of multiple, adjacent frames. As a result, the proposed algorithm can remove interpolation artifacts based on a degradation model of low-resolution images. The dictionary is classified by the orientation of patches for fast searching. The proposed algorithm consists of two steps: i) dictionary generation based on the image degradation model and ii) multi-frame image reconstruction for super-resolution. Experimental results show that the proposed SR algorithm provides better reconstructed images with less undesired artifacts than existing methods.