摘要
当前很多视频对象分割方法都联合利用了多种特征进行前景提取,但是这些特征都是通过简单加权融合到一起的.该文通过主成分分析法(Principal Component Analysis,PCA)比较准确地衡量了各特征在前景检测中所占的权重,使其有效指导前景分割.同时通过对各特征建立相应的高斯模型,有效提高了前景分割的质量,最后再通过基于颜色不变量的阴影检测算法得到了比较准确的结果.实验中采用了颜色(RGB)和局部二值模式(Local Binary Pattern,LBP)4种特征,结果表明:无论是对于静态场景还是动态场景,该算法都具有良好的分割效果.
Now there are many video object segmentation methods combined using many fea- tures. However, these features are fused together through the sample weighted. We measured every feature＇s weight by the foreground detection, make it effective for foreground segmenta- tion. Additionally, Gaussian model is built for each feature, which improved the quality of seg- mentation effectively. Then more accurate results are obtained with the shadow detection method based on the invariant color features. We use the RGB color features and local binary pattern （LBP） in our experiment. Experiments on videos demonstrate the efficiency of our proposed method.
出处
《计算机学报》
EI
CSCD
北大核心
2013年第11期2356-2363,共8页
Chinese Journal of Computers
基金
国家自然科学基金(61379106)
山东省自然科学基金(ZR2009GL014)
山东省中青年科学家奖励基金(BS2010DX037)
国家文化部科技创新基金(46-2010)
中央高校基本科研基金(09CX04044A
10CX04043A
10CX04014B)资助~~
关键词
主成分分析
颜色
局部二值模式
高斯模型
颜色不变量
principal component analysis
RGB
local binary pattern
Gaussian model
invariantcolor features
