基于分水岭-隐马尔可夫模型联合的语义视频对象提取技术  

SEMANTIC VIDEO OBJECT EXTRACTION BY HYBRID WATERSHED-HIDDEN MARKOV RANDOM FIELD MODEL

作　　者：杨树堂[1] 汤志平[1] 

机构地区：[1]上海交通大学电子信息与电子工程学院,上海200030

出　　处：《计算机应用与软件》2007年第11期9-11,共3页Computer Applications and Software

基　　金：国家高科技研究发展技术(863)项目(2005AA145110)资助

摘　　要：针对3C无线网络的增值服务业务,提出一种语义对象基于分水岭算法的视频帧内的分割,及基于隐马尔可夫模型的视频帧间跟踪提取技术。其主要的特点是首先采用基于标识集的分水岭算法来进行初始帧内语义视频对象的标定和分割处理,随后再进行二值化掩膜处理,最后借助隐马尔可夫测量场模型,将后续帧中视频对象的跟踪处理演化为跟踪区域与非跟踪区域的二值离散化标定问题。实验结果证明,该算法能很好地实现视频帧序列中语义视频对象的连续提取。In order to deal with the increment service of 3G wireless network, a semantic video object extracting technique based on hybrid watershed-hidden Markov random field model is put forward. Its main characteristic is that the semantic video object in the initial frame is identified and segmented by improved watershed algorithm based on identifier sets, and then the SVO is masked by two valuse 0 and 1. Finally the video object is tracked in the next frames by using the hidden Markov Random field model. The tracking process can be regarded as discrete labeling problem. The results of the experiment show that the algorithm performs well in semantic video object extraction.

关 键 词：语义视频对象 分水岭算法 隐马尔可夫模型 极低比特率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]