【作者】 陈博；

【导师】 王保保；

【作者基本信息】 西安电子科技大学 ， 计算机应用技术， 2004， 硕士

【摘要】 本文以视频对象分割技术为研究课题，首先介绍视频分割相关的理论与技术，然后对现有的基于运动和基于时空域相关两大类分割算法进行对比研究，并把重点放在基于3D区域生长的时空域分割算法的分析上。 从时空域分割要着重解决的几个关键问题入手，本文探讨了3D区域生长的种子分布和生成方法，给出了区域生长过程中的像素和元素之间的相似度准则和后处理过程，并构建了相应的时空域数据结构来支持生长算法的进行。通过区域生长算法，输出视频中具有颜色同质性的组件，接下来进行运动估计和元素运动轨迹分析得到这些同质组件的运动信息，并用空间聚类算法将具有运动一致性的组件合成视频对象。此外，本文还对视频时域分割、颜色空间选取、空域滤波等时空域分割要解决关键问题进行了探讨，并提出了一种自适应阈值切变镜头探测算法和加权中值滤波算法来解决这些问题。最后，将上述算法结合起来形成一个视频对象分割方案，有效地解决运动前景和背景分离的问题，并成功地完成从视频图像序列中抽取视频对象板的任务。更多还原

【Abstract】 Video object segmentation techniques are discussed, theories and techniques related to video segmentation are introduced and the existing typical algorithms of motion-based and spatiotemporal segmentation are analyzed and compared with the emphasis on analysis of spatiotemporal segmentation algorithms based on 3D region growing.Proceeding with several key problems about spatiotemporal segmentation, this paper discusses the generation and distribution of seeds in 3D region growing, provide the similarity measurement between pixel and volume, design the post processing and construct spatiotemporal data structure to support the algorithm. Homogeneous video components with similar color feature are obtained. Their motion trajectory is analyzed and motion estimation is made, and these components are clustered into objects with motion coherence. In addition, other key problems such as video temporal segmentation, color space selection and temporal filtering are discussed and an adaptive threshold video shot cut detection algorithm and a weighted median-filtering algorithm are presented as solution. At last, the algorithms are combined into an automatic video object segmentation schema, which can separate motion foreground from stationary background and extract video object plane from video image sequence in succeed.更多还原