【摘要】 保持时域连贯性与保证图像空域的重建质量在视频重定向中至关重要。针对这个问题，提出一种基于运动历史图的网格变形方法进行视频大小的调整。通过估计帧间摄像机的运动，将视频帧对齐在时间轴上，构造全局运动模型，从而获得全局运动历史图；利用运动历史图来顺序指导单帧图像形变，将单帧图像的放缩问题转换为网格变形问题，通过构造一系列约束条件在网格上建立保持整体形状的最优化模型，实现背景的时间连续性的同时又有效地保持运动对象的形状。测试结果显示无论是处理静止的视频还是具有复杂运动的视频，该方法都能获得令人满意的结果，且能实现对视频流的实时处理。更多还原

【Abstract】 It is very important to maintain temporal coherence and ensure the quality of image spatial reconstruction in video retargeting. To solve this problem, this paper proposes the grid deformation method based on motion history image to adjust the video size. By estimating the motion of the inter-frame camera, the video frames were aligned on the timeline, and the global motion model was constructed to obtain the global motion history image. We used the motion history image to guide the deformation of single frame image sequentially, and transformed the contraction problem of single frame image into the grid deformation problem. By constructing a series of constraints, an optimization model to maintain the overall shape was established on the grid to achieve the temporal continuity of the background and effectively maintain the shape of the moving objects. The experimental results show that this method can obtain satisfactory results and process video stream in real time for both stationary and complex motion videos.更多还原