摘要
为了更好地利用单演幅值和区域主方向信息,分别提出了一种单演韦伯差异激励局部块二值模式和单演区域主方向模式,并在此基础上进一步采用分块子模式策略融合两种特征。该方法首先对单演幅值求取差异激励,将差异激励分解为正值和幅值图像;然后对正值和幅值图像采用基于分块的局部二值模式编码,采用主成分分析方法求取单演区域主方向,并对主方向进行均匀量化,再采用异或编码。在获取两种特征后,采用分块子模式的策略对两种特征进行加权融合。在AR和CAS-PEAL上的实验表明,MWLMBP和MDOP两种特征提取方法能够有效提取图像的判别信息,进一步融合两种特征的方法能够有效增强特征的分类能力,提高特征的识别性能。
In order to make full use of monogenic magnitude and orientation, this paper proposed a method based on monogenic magnitude Weber difference excitation local binary pattern and monogenic region domain orientationg And it proposed a method which fusion the two patterns using block-based sub-patterns. Firstly, it calculated the monogenic magnitude Weber difference ex- citation, and divided the excitation into positive image and negative image, after that, used the muhi-blocks local binary patterns to encode both images. Secondly, it calculated the region domain orientation with principal component analysis method, and then quantized it uniformly and encoded it with XOR method. After obtained the two features, it used the block-based sub-patterns strategy to fusion them with weight. The experiments on AR and CAS-PEAL show that, monogenic weber excitation local binary patterns and monogenic domain orientation patterns can extract discriminate information effectively, and the fusion with the two features can enhance the classification ability of features, and improve the recognition performs.
出处
《计算机应用研究》
CSCD
北大核心
2014年第4期1246-1251,共6页
Application Research of Computers
关键词
单演信号
韦伯
局部二值
主成分分析
区域主方向
monogenic signal
Weber
local binary pattern
principal component analysis
region domain orientation
