【摘要】 给出一种基于全局运动仿射模型的视频目标移除方法.用仿射模型描述非移除区域的全局运动,通过提取相邻帧的匹配对进行模型参数估计;对于每一帧中待修补的移除区域,根据仿射模型之间的递推关系和运动一致性准则,从其它帧中寻找该区域的信息并进行修补.实验表明,方法能够较好的对运动前景、台标等目标进行移除.更多还原

【Abstract】 A video object removal method using global motion based affine model is proposed.Non-removal region which is usually regarded as global motion is described by affine model.And then,matching points from two continuous frames are extracted to estimate affine model parameters.By taking account of the recursion of affine model and the global motion coherence,the removal region in each frame is repaired according to the corresponding information of other frames.The experimental results show that the proposed method achieves a better performance in removing objects,such as foreground,logo and so on,than some other methods.更多还原