一种基于局部模型对齐的动态视频修复方法

LOCAL MODEL ALIGNMENT FOR DYNAMIC VIDEO INPAINTING

传统的动态视频修复问题中,全局模型只用于定点拍摄或者场景是平面的情形,当摄像机运动范围较大时,视差大会导致视频修复结果出现模糊和重影等现象。针对此问题,将局部模型引入视频对齐方法中,主要思想是将视频帧的局部区域近似为一个小平面,局部区域用一个Homograph描述,对不同区域采用不同的局部模型。在局部模型的估计中融合了基于运动一致性的对应点匹配方法,增加了特征点的匹配数,从而提高局部模型的对齐精度。实验结果表明,该方法大大提高了视频对齐精度,从而有效地改善了视频修复性能,扩大了传统视频修复方法的适用范围。

In the traditional dynamic video repair problem,the global model is only used in the case of fixed-point shooting or in the case that the scene is planar.When the motion range of the camera is large,the parallax results in blur and ghost of the video repair results.In view of this problem,this paper introduces the local model into the video alignment method.We took the local region of the video frame as a small plane.The local region was described by a homograph,and different local models were used for different regions.We fused the corresponding point matching method based on motion consistency in the estimation of the local model,and increased the matching number of feature points,so as to improve the alignment accuracy of the local model.The experimental results show that the proposed method greatly improves the accuracy of video alignment,thus effectively improving the video repair performance and expanding the scope of application of traditional video repair methods.