特征提取是人脸识别中一个关键步骤。传统的Fisherface人脸识别方法中用样本的类均值和总体均值定义相应的散布矩阵,丢失了样本个体之间的结构信息,本文提出了一种基于原始样本个体结构信息的结构化Fisherface人脸识别方法,最后得到的特征数据中保留了原始样本更多的分布信息。在ORL人脸数据库的实验结果验证了该方法的有效性。
The feature extraction is one of the key mean are used to define the corresponding scatter steps in the face recognition. The class mean and the total matrices in conventional Fisherface method with which the structure information between samples is discarded. A new feature extraction method named structurized Fisherface is proposed. More distribution information of original samples is preserved in the final feature space. Experimental results from the ORL face database proved the effectiveness of the proposed method.