| 题名 | 数字视频自动分割技术及其应用研究 |
| 作者 | 高丽 |
| 学位类别 | 博士 |
| 答辩日期 | 2007-06-08 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 声学研究所 |
| 关键词 | 智能视频处理 视频镜头分割 视频物体分割 图像分割 成份分析 Watershed图像分割 变化检测模板 数学形态学 |
| 其他题名 | Theory and Implement Study on Digital Video Automatic Segmentation and Application |
| 学位专业 | 信号与信息处理 |
| 中文摘要 | 智能视频处理技术涉及到许多基本技术，由于存在一些关键性技术仍然没有得到很好地解决，致使智能视频处理技术在其发展过程中受到一定程度的阻碍。本论文针对当今各种智能视频处理技术进行了研究与总结，发掘其中存在的一些关键技术问题。其中，数字视频自动分割技术是智能视频处理技术的基础性与关键性技术之一，同时，它也是MPEG-4和MPEG-7标准中基本与重要的组成部分。本论文围绕数字视频自动分割技术展开深入研究，期望寻找新的算法设计思路，引入新的理论知识，这涉及到信号处理、模式识别和图像处理等理论领域，从而进一步提高数字视频自动分割技术的性能。从上至下，数字视频分割技术主要由视频镜头分割、多帧图像的视频物体分割和单帧图像分割三个部分组成。 当今，面对海量的视频数据以及有限的网络资源和存储资源，网络传输和存储技术都已经采用压缩的视频形式。完全解压缩一方面增加处理时间，另一方面耗费系统资源。越来越多的研究学者开始研究开发直接在压缩域中的视频处理技术，目的是提高算法效率，节省系统资源。考虑到压缩域视频特征提取在本论文后续的研究工作和本次项目课题中都是一个研究基础，本论文首先仔细研究了MPEG压缩域的视频特征提取算法，并且就几个典型算法进行了深入研究与介绍。 视频序列中蕴含着十分丰富的信息，如何从如此庞大的信息量中提取有用的视频特征，对于视频镜头分割技术起着举足轻重的作用。至今仍然没有十分有效的算法来解决镜头分割问题，特别是渐变镜头分割问题，其中一个重要原因就是没有寻找到一种合适的视频特征可以有效表征渐变镜头期间的不相似性，而同时还能够有效抑制快速摄像机运动和物体运动带来的干扰。本论文从两个方面入手，尝试提高镜头分割性能。本论文首先提出将模式识别理论中的成份分析方法引入到镜头分割技术中，用来提取视频特征。理论与实验证明，利用本论文提出的方法提取得到的视频特征在诸多方面优于传统的镜头分割特征，可以有效表征渐变镜头期间的不相似性，同时还能够有效抑制摄像机运动和物体运动带来的干扰，对噪声的敏感度明显下降。另一方面，考虑到镜头分割算法的设计同样很重要，本论文结合前面提取的视频特征设计出一种联合突变镜头和渐变镜头的分割算法，实验结果表明，本论文的镜头分割算法性能得到有效提高，特别是对渐变镜头分割问题的贡献比较大。 针对视频物体分割技术，本论文选择了其中一种目前比较流行的物体分割思路，即基于变化的检测模板，展开深入研究。针对该思路下分割方法存在的几个技术问题，本论文引入新的设计思路和理论方法尝试解决这些问题。本论文首先针对物体和背景边界强度较弱时而引起的分割丢失问题，提出在预处理阶段对灰度图像进行局部对比度增强处理，然后用对比度增强后的图像参与算法的计算。通过设计3 3模板的滤波器滤除对比度增强带来的噪声问题。当视频序列的背景复杂并且不能够从首帧中获取背景信息时存在的问题，本论文提出利用随机信号的统计特性积累得到当前帧的背景信息，并用其参与后续算法的计算，实现自动获取所需的背景信息。同时算法提出背景累积过程中分类讨论的观点，很好地解决了运动物体停止运动时间较长而出现丢失的现象。 本论文中针对单帧图像分割算法进行了深入研究，期望通过新的解决思路更好地推进图像分割技术的发展。本文选择了一种非常著名的，比较有应用前途的图像分割方法——Watershed算法，对其存在的过分割问题进行了深入分析。我们提出了一种新颖的标记提取方法，受到图像处理中的照度-反射模型的启发，提出了从图像梯度的低频成份中提取与图像中各个物体相关的极小值，构成标记图像。利用形态学中的极小值标定技术修改原始梯度图像。在修改后的梯度图像上进行Watershed图像分割。理论与实验证明，本章提出的新思路可以有效缓解传统方法中简化图像与保护边缘的矛盾体，同时也使形态学中的H-minima技术性能得到有效提高。通过对不同类型的图像进行试验，证明本章提出的图像分割算法能够获得符合人类视觉特点，具有实际意义而且均一的分割区域，以及较为准确、连续、一个像素大小的物体边界。与其他的 Watershed改进方法相比，本文提出的方法要求的计算复杂度较低，具有简单的参数，同时能够更为有效地降低Watershed算法的过分割问题。 |
| 英文摘要 | The research of video processing focus on intelligent implement, while the technique and multimedia application is developing faster and faster. Intelligent video implements include content-based video retieval, video abstract, video analysis, video annotation, and video management etc. As the basis and core problem of these issues, digital video automatic segmentation is attracted attention by more and more researchers. Furthermore, the derivation of the development of Internet and communication, quite a large amount of compressed bit streams, make researchers pay more attention to extract video and image features in compressed domain. It can make video and image processing faster by avoiding decompression process. In this thesis, in order to improve the performance, automatic video segmentation approach is studied deeply, and new idea, new theories, including signal processing, pattern recognition, image processing etc. are applied to the video segmentation. Up to down, the video segmentation can be mainly constitute of three parts: video shot segmentation, video object segmentation, image segmentation. Since the original data are compressed to preserve storage space, and decompression is extremely costly, the algorithms designed to operate with multimedia data should ideally do so in the compressed (or minimally-decoded) domain. Video processing directly in the compressed is one basis of this thesis, therefore we firstly study the method about video feature extraction from the compressed domain, and introduce several well-known related algorithms detailed. An automatic, real-time detection approach to video scene change detection is presented. Owing to the high correlation of two consecutive video frames, it is proposed that only the eigenvector corresponding to the largest eigenvalue is retained in the principal component analysis (PCA) for video data. A one-dimensional PCA feature of video data is then generated from the PCA. It shows superior performance compared to the histogram feature and the pixel feature. The detection algorithm based on this PCA feature is then designed to detect both abrupt and gradual transitions. The proposed approach is tested on the TREC video test repository to validate its performance. On the basis of combining change-detection-based segmentation approach and spatial edge information by canny edge detection, an algorithm is proposed in which local contrast enhancement; then for the complex background , the algorithm utilizes probability-based classification to accumulate the background information, which it is needed by the original segmentation algorithm, and consequently realizes the capturing of background information automatically; Finally, the paper proposed that three situation should be discussed in the process of accumulating background information. The proposed algorithm is evaluated on several MPEG-4 test sequences and produces promising results. This paper proposes a novel automatic image segmentation system, which improves the traditional marker-based watershed transform. A new marker-extracted approach is proposed to extract the regional minima from the low frequency components in the gradient map. The extracted minima constitute the binary marker image. Then the original gradient map is modified by suppressing its all-intrinsic minima around these extracted markers. Thus, compared with the traditional approach, both the spurious minima are more effectively removed and meanwhile the boundaries of objects are more effectively protected. Additionally, the performance of H-minima technology in our approach is improved further than the traditional application of this technology. Finally, the Watershed algorithm is applied to the modified gradients to perform the segmentation. Across a variety of image types, it is proven that this new system can obtain meaningful and homogeneous regions with accurate, consecutive and one-pixel wide boundary. Additionally compared with other methods, this system requires fewer computations and simpler parameters and can more efficiently reduce the oversegmentation of the watershed algorithm. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 133 |
| 内容类型 | 学位论文 |
| 源URL | [http://159.226.59.140/handle/311008/192]  |
| 专题 | 声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 高丽. 数字视频自动分割技术及其应用研究[D]. 声学研究所. 中国科学院声学研究所. 2007. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论