CASIA OpenIR  > 毕业生  > 硕士学位论文
基于多尺度数学形态学的彩色视频目标分割
其他题名Color Video Object Segmentation Based on Multiscale Morphology
赵政文
学位类型工学硕士
导师卢汉清
2002-05-01
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词多尺度 目标分割
摘要随着MPEG-4和MPEG-7的研究和发展,以及近几年数字视频图书馆技术 的崛起,基于内容的编码和面向对象的存取和操作技术同益得到人们的重视,视 频分割技术迅速成为当前视频研究领域的热点。视频分割是新。代视频编码、视 。频检索、基于互联网的多媒体交互等新兴领域的关键技术。本文首先介绍了基于 数学形态学的多尺度图象分割方法,然后对鲁棒性运动估计进行丁讨论,结合前 两部分的工作,最后给出了一个时空域运动估计和分割的框架。 数学形态学是对图象进行非线性处理的一种工具,它首先应用于二值图象, 然后经扩充应用于灰度图象,近年来,人们试图在彩色图象中引入矢量偏序的关 系,从而用形态学处理彩色图象。本文把形念学和矢量数据处理的方'法结合起来, 首先用矢量中值滤波器去除图象中的噪声,在对图象进行空域分割的同时,按照 分割尺度的大小建立了区域金字塔,并且对于水线分割的过分问题,提出了一种 多层次的处理方法,保证了分割区域的有效性,另一方面,采用层化变换解决了 水线漂移问题,建立了严格的多尺度处理的尺度空间,为后面的运功分割奠定了 很好的基础。 为了对视频序列中的运动进行准确的估计,我们在鲁棒性运动模型参数估计 框架下,使用了一种凸M估计子,与传统的M估计子相比,它在汁算效率和稳 定性方面都具有一定的优势. 针对视频目标分割的任务,我们采用一种以运动一致性为准则的区域分裂和 基于复合测度区域增长为空间拓扑约束的多运动估计和分割方法,在一定程度上 解决了运动分割中的遮挡和形变等问题,结合人机交互标定视频对象的工作,形 成了一个完整的半自动视频目标分割的框架,在实验中取得了较好的结果.
其他摘要With the development of MPEG-4, MPEG-7 and the progress of the digital video library, much research efforts have been focused on the methods of content-based video coding and object-based multimedia access and manipulation. The technique and technology of video segmentation become the focus of the domain of the video research. Video segmentation is a key technology in the new generation of the video coding, video indexing, Internet-based multimedia interacting and etc. In the thesis we firstly present a algorithm of multi-scale image segmentation based on mathematical morphology. And then the thesis discusses how to get the robust motion estimation. Combined with the above work, we conclude a framework of spatial-temporal motion estimation and segmentation. Mathematical morphology was originally applied to the binary image as a nonlinear signal processing tool. Subsequently it was extended to grayscale image processing. Lately the researchers try to process the color image using mathematical morphology by defining a vectorial ordering relation that induces a lattice structure on the data. In contrast to the above method, the thesis employs the algorithm of combining mathematical morphology and vectorial data processing. Above all. the vector median filter performs a nonlinear filtering operation in order to suppress the impulsive noise. While performing image segmentation, we compose a region pyramid according to the different scales.' As for the serious over-segmentation produced by the classic watershed algorithm, we propose a multiplayer measure to guarantee the validity of the regions. On the other hand, we use a stratification transformation to solve the problem of feature deviation. In this way, we construct a scale-space of multi-scale image processing, which gives a sound foundation for the following work about video object segmentation and tracking. For the sake of getting the accurate motion estimation, we use a convex M-estimator under the framework of robust motion parameter estimation. In contrast to the classic M-estimator, it has some advantages in view of the computational efficiency and stability. In order to accomplish the video object segmentation, we employ a multiple motion estimation and segmentation method, which uses the region spill based on motion consistency and the region growing based on multiple criteria :~s the spatial topological constraints. In a way, it solves the problems of occlusion and deformation in motion segmentation. Combined with the work of the interactive definition of the video objects, we present a complete framework of semiautomatic video object segmentation.
馆藏号XWLW628
其他标识符628
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6863
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
赵政文. 基于多尺度数学形态学的彩色视频目标分割[D]. 中国科学院自动化研究所. 中国科学院研究生院,2002.
条目包含的文件
条目无相关文件。
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[赵政文]的文章
百度学术
百度学术中相似的文章
[赵政文]的文章
必应学术
必应学术中相似的文章
[赵政文]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

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。