一种多层次语义视频对象分割算法  

A New Algorithm of Multi-hierarchy Semantic Video Object Segmentation

作　　者：汤志平[1] 杨树堂[1] 李建华[1] 

机构地区：[1]上海交通大学电子工程系,上海200030

出　　处：《上海交通大学学报》2007年第1期15-18,共4页Journal of Shanghai Jiaotong University

基　　金：国家高科技研究发展技术(863)项目(2005AA145110)

摘　　要：提出一种基于高斯马尔可夫随机场及规则化图划分的多层次语义视频对象分割算法,其主要特点是将视频序列帧中对象的分割看成是“内容树”结构中复合结点的形成过程.首先使用高斯-马尔可夫模型来进行视频帧内的最优标记场标定,然后引入规则化图划分准则进行过分割区域的合并,得到具有语义意义的视频对象.实验表明,本分割算法具有较高的准确性,误差的均值为11.375%,标准方差为0.94%.A new algorithm of multi-hierarchy semantic video object segmentation based on Gaussian-Markov random field and normalized graph partition was presented. Its main character is looking the segmentation of video sequence frames as the formation of compound nodes in the content tree structure. Firstly, pixels of intra-frame are labeled through optimizing the energy function derived from the Gaussian-Markov random field model. In order to merge the over-segmented regions and get the semantic video objects, the criterion of normalized partition in graph theory is introduced. In the experiment, error mean is 11. 375% and standard variance is 0.94%, which shows it is an efficient algorithm to segment the semantic video object.

关 键 词：语义视频对象 高斯-马尔可夫模型 规则化图划分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]