基于二维局部鉴别高斯的特征提取方法  

Feature Extraction Based on 2D Local Discriminative Gaussians

作　　者：张智斌[1,2] 朱俊勇[1] 郑伟诗[3] 王倩[1] 赖剑煌[3] 

机构地区：[1]中山大学数学与计算科学学院,广州510275 [2]华南理工大学数学系,广州510641 [3]中山大学信息科学与技术学院,广州510275

出　　处：《计算机科学》2014年第6期275-277,共3页Computer Science

基　　金：国家自然科学基金(61128009);国家科技支撑项目(2012BAK16B06);广东省科技计划项目(2012B010100035);广东省自然科学基金(S2012010009926);中央高校基本科研业务费专项资金(2013ZM0094;2013ZM0114)资助

摘　　要：特征提取是人脸识别的关键。特征提取方法一般需要预先把二维图像转化成一维图像向量。然而高维的图像向量会导致不能快速、精确地计算所需的协方差矩阵及其特征向量。针对该问题,提出了一种基于二维局部鉴别高斯的特征提取方法(2D-LDG)。该方法继承一维局部鉴别高斯降维方法的优点,其目标函数是留一交叉验证误差的光滑逼近,并且只考虑训练样本的局部分布,对训练样本的全局分布不做任何假设。同时,2D-LDG直接对二维图像做特征提取,不需要事先把图像转化为维数巨大的图像向量,能快速、精确地计算协方差矩阵及其特征向量。在ORL、YaleB人脸数据库上的实验结果表明,2D-LDG特征提取方法有良好的识别效果。Feature extraction plays an important role in face recognition.In general,feature extraction methods need to transfer the 2D images into 1D vectors.As a result,it is hard to calculate the covariant matrix and eigen-vector efficiently and exactly due to the high dimensionality.This paper proposed a new feature extraction method named 2D local discriminant Gaussian （2D-LDG）.It inherits the properties of LDG and the objective function of proposed method is also an approximation to the leave-one-out training error of a local quadratic discriminant analysis classifier.Also,it applies local Gaussians to eastimate probability in each point,relaxing the assumption on the class probability density function.Meanwhile,2D-LDG is operated on 2D images directly which avoids turning the image matrixes into high dimensional vectors,and is able to calculate the covariance matrix and eigen-vector in a more efficient and accurate way.Experiments on ORL and YaleB-Extended show that our proposed 2D-LDG feature extraction method achieves better performance in face recognition.

关 键 词：特征提取 局部鉴别高斯模型 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]