基于时空一致的三维双向相似性优化的视频编辑
首发时间:2012-07-24
摘要:本文基于时空一致的三维双向性优化纹理合成方法,构建了一个基于用户交互的集视频缩放、视频修复、视频洗牌等多种编辑为一体的统一视频编辑系统。为了保证编辑得到的结果与原始视频之间的连续性、完整性,我们首先提出了一种针对视频数据的三维双向相似性度量方法,将视频编辑转化为一个基于三维双向相似性度量的块优化问题。更进一步,通过在三维双向相似性方程中加入一个一致项,我们有效地解决了合成后视频抖动、细节丢失等问题。本文采用最大期望值方法迭代优化求解该优化模型,并在全局优化的过程中构建原始视频和目标视频的各个信息通道的直方图匹配算法,进一步提高优化后视频的质量。实验室结果表明,本文提出的视频编辑系统不仅能够有效支持视频缩放、视频修复及视频洗牌等编辑处理,而且较好的解决了视频合成算法中存在的特征断裂、视频抖动、细节丢失等问题。
关键词: 计算机应用 视频编辑 双向相似性 最大期望值算法 直方图匹配
For information in English, please click here
Efficient Video Editing System Using 3D Bidirectional Similarity
Abstract:We construct a unified interactive video editing system support video retargeting, video completion, video reshuffling and some other editing base on an optimized texture synthesis method using space-time consistent 3D bidirectional similarity measure. In order to ensure coherence, completeness between the output video and original video, we first proposed a 3D bidirectional similarity measure, transform video editing into a 3D bidirectional similarity measure based patch optimization problem. Further, by adding a consistent term in the 3D bidirectional similarity equation, we effectively avoid jittering and detail losing during the synthesis. In this paper, we use the expectation maximization to iteratively optimize the model, we further improve the video quality by applying color histogram matching during the optimization. Experiment results show that the proposed video editing system can not only effectively support video retargeting, video completion, video reshuffling and so on, but also avoid issues like texture broken, video jittering, detail losing.
Keywords: computer application video editing bidirectional similarity expectation maximization histogram matching
论文图表:
引用
No.****
同行评议
共计0人参与
勘误表
基于时空一致的三维双向相似性优化的视频编辑
评论
全部评论