基于DCT和MMC的人脸识别方法  

Face recognition method based on DCT and MMC

作　　者：韩晓翠[1] 

机构地区：[1]临沂师范学院信息学院,山东临沂276000

出　　处：《计算机工程与设计》2010年第14期3284-3286,3320,共4页Computer Engineering and Design

摘　　要：提出了一种基于离散余弦变换(DCT)和最大间距准则(MMC)鉴别分析的人脸识别方法。先对人脸图像进行离散余弦变换,选取变换矩阵左上角的一部分变换系数构成特征向量,然后对所有训练样本按照最大间距准则鉴别分析算法计算投影矩阵,把人脸图像矩阵在投影矩阵上投影得到特征矩阵。融合决策阶段,在以上两类特征集中,基于欧氏距离测度分别计算待识别样本到所有训练样本的距离并对得到的两类结果采用加权方法进行融合,得到最终的分类结果。基于ORL人脸数据库的实验结果表明了该算法的有效性。A face recognition algorithm based on discrete cosine transformation （DCT） and maximum margin criterion （MMC） discriminant analysis is proposed. Firstly,discrete cosine transformation is performed on face image matrix,and some transformation coefficients which locate in top left corner of transformed matrix are selected to form feature vector. Secondly,according to maximum margin criterion,projection matrix of all training samples is calculated,so a feature matrix can be obtained when projecting a face image on it. During fusion period,Euclidean distances between training samples and testing sample are calculated respectively in these two kinds feature sets. Subsequently,weight fusion operation is done on these results that are gotten. Finally,the nearest distance classification is used to distinguish each testing face sample. The results of several experiments that performed on ORL face database indicate that the proposed approach is effective.

关 键 词：人脸识别 离散余弦变换 最大间距准则 投影矩阵 加权融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]