Abstract: A new super resolution reconstruction approach on periodic video is proposed. It reconstructs one period clear video from a long video with several periods. Firstly, it aligns the video by the detected periods and groups the frames. Then each high resolution image in one period can be reconstructed by the frames which the object poses in the same position. For enhancing image resolution from multi-frames, it proposes a direct computation method which is different from the iterative model based approaches such as back-projection approaches etc. which had been proposed. Experiments on simulation images and real images show that the approach is better than other model based super resolution methods.

Key words: super resolution, periodic video, mutual information

摘要： 通过重建周期视频一个周期内的每一帧从而恢复出一个周期的高质量视频。由于视频是低帧率的，相邻帧间差别较大，无法使用相邻的帧进行基于模型的超分辨率重建，于是提出了配准视频周期，然后利用出每个周期内的相同景物的帧进行超分辨率重建，从而为低帧率低分辨率单个周期视频找到了超分辨率重建的方法；在帧间配准方面使用了互信息方法，取得了更精确的效果；使用的超分辨率重建方法无需迭代，对图像直接进行重建，算法更为简单。在仿真实验和实际实验中均取得了比其他基于模型的方法更好的结果。

关键词: 超分辨率, 周期运动视频, 互信息