基于局部定性表达的光照不变人脸识别方法  

Illumination invariant face recognition algorithm based on local qualitative representation

作　　者：张熠[1] 郭琳[2] 张桂林[1] 

机构地区：[1]华中科技大学图像识别与人工智能研究所,武汉430074 [2]武汉大学电子信息学院,武汉430079

出　　处：《计算机应用》2008年第5期1276-1279,共4页journal of Computer Applications

摘　　要：提出一种局部定性特征表达方法称为局部序映射(LOM)作为光照不变表达用于人脸识别。该特征表达像素在其邻域中所处的序阶位置,反应了图像像素在局部区域内的相对关系,因而对光照引起的图像大面积灰度变化有很好的不敏感性。同时提出一种针对序定性特征的逻辑距离度量,用于特征空间中最近邻分类器的相似性度量准则。在Extended YaleB光照变化人脸测试库中的实验表明了本文局部序映射表达的光照鲁棒性及逻辑距离度量的有效性。An illumination invariant face recognition algorithm based on local qualitative representation called Local order mapping was proposed. This representation encoded the order rank of the central pixel among its neighbors, and represented the relative configuration of the image＇s local regions. Therefore, the local order mapping was very insensitive to the large intensity variations caused by the illumination changes. Meanwhile, a new logic distance measurement for ordinal qualitative features was applied in the nearest neighbor classification for recognition. The experimental results on Extended YaleB face dataset show the illumination robustness of the proposed representation and the validity of the logic distance measurement.

关 键 词：序特征 光照不变 人脸识别 相似度量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]