Abstract: Aiming at video inpainting when the camera moves parallelly to the scene and shakes，an algorithm that transforms video inpainting under complicated camera motion to video inpainting under stationary camera is proposed.First，a new video that contains stationary background is got by a new mosaic method based on the algorithm of SURF，then foreground and background inpainting are separately carried out in the new video，lastly，the inpainted original video is got by mapping the inpainted new one to the original.In foreground inpainting，the target patch is designed as a rectangle of size that can cover the region of the entire foreground.In background inpainting，marking and inpainting of the region under repair is done uniformly.Experiment results show that this algorithm enhances accuracy and speed and favors the continuity of the video.

Key words: video inpainting, image registration, image mosaic, target patch

摘要： 针对摄像机平行于场景运动，且存在晃动的情况，提出将摄像机运动情况下的视频修复转化为摄像机静止情况下的视频修复。该算法采用基于SURF特征点提取的视频拼接技术，将输入视频转化为新的背景静止的视频序列，在新视频中分别进行前景修复和背景修复，把修复好的新视频映射到原视频。在前景修复中，将目标块定义为能够覆盖整个运动物体的长方形块。在背景修复中，算法对待修复区域进行统一的标记和修复。实验结果表明，该算法提高了修复的准确度和速度，有利于保持视频的连续性。

关键词: 视频修复, 图像配准, 图像拼接, 目标块