基于匹配帧的层次化结构的视频分割技术-计算机科学与技术专业论文.docx

第 第 PAGE IV 页 第 第 III 页 首先，本文总结和比较了前人对于图像层次化分割的研究，通过实 验分析发现，迭代 mean-shift 算法虽然原理简单，但效果不佳，基于鞍 点检测的算法虽然有很好的理论基础，但运行时间较长。本文最终选取 了一种基于鞍点检测的近似算法，该算法既有很好的理论基础，又有着 不错的分割结果。 其次，本文在图像层次化分割算法的基础上提出了将其应用于视频 分割的算法。该算法的想法是源于对视频分割中的不一致性的原因的研 究。我们通过研究发现，基于空域优先的视频分割算法的不一致性并非 由于分割算法本身造成的，而是由于在寻找帧间关联的时候没有考虑到 图像的层次化信息。本文对帧间关联问题进行了抽象，将寻找帧与帧关 联的问题转化为树与树之间相似度的问题，由于该问题的难解性，我们 寻找了一种近似的求解算法，即树的最大覆盖的问题，并给出了解决方 案。 此外，本文设计了一个交互式视频分割系统原型，并在人机交互方 面有所创新，在关键帧标注方面，我们提出了应用图像的层次化信息来 进行逐层标注。 关键词：视频分割; 层次化结构; 带权树的覆盖; 树与树的相似度 VIDEO SEGMENTATION BY MATCHING HIERARCHY STRUCTURES OF FRAMES ABSTRACT Owing to the fast development of computer techniques and the digital media industry, the video processing techniques become gradually a hot research topic in the fields of digital image processing and digital media. Video segmentation technique is an important role in video processing techniques such as video NPR, video inpainting, video structure analysis and so on. The objective of video segmentation is to extract some meaningful video objects from video sequences. It develops from the image segmentation and can be viewed as an extension of image segmentation. Compared with image segmentation techniques, video segmentation has larger processing data, and should keep consistent in time dimension. Image hierarchical segmentation develops from the image segmentation techniques. When we do a hierarchical segmentation, we can either do a coarse segmentation, dividing the image into a few big parts, or do a fine segmentation, dividing the image into lots of parts. We can construct the hierarchical segmentation result from these segmentation results. Generally, a hierarchical segmentation is desirable, and is similar to our habit for observing objects, from an overview to the details. This paper focuses on how to use hierarchical structures to improve current video segmentation techniques. We propose a novel video segmentation method using the hierarchical structure of each frame. Based on this method, we d