Abstract: The method based on linear projection is in the present one of important mainstream methods，has obtained widespread attention and made remarkable progress in the recent years.One-dimensional linear projection methods includes Eigenface method and Fisherface method and so on.Two-dimensional linear projection methods includes two-dimensional principal component analysis and two-dimensional linear discriminant analysis，as well as a series of their expanded methods and so on.On this basis，a novel two-dimensional matrix feature extraction method is proposed in this correspondence.The experiments on ORL standard face-database indicate that this method has a higher recognition accuracy and needs less computational and memory requirements than other existing methods.

Key words: two-dimensional principal component analysis, bidirectional linear discriminant analysis, feature extraction, principal component analysis, linear discriminant analysis

摘要： 基于线性投影的方法是目前人脸识别领域中重要的主流方法之一，在近年中得到了广泛的关注，取得了显著的发展。其中，基于一维线性投影的方法包括特征脸方法和Fisher脸方法等；基于二维线性投影的方法包括二维主成分分析和二维线性判别分析，以及它们的一系列拓展算法等。在此基础上，给出了一种基于二维矩阵的特征提取新方法。通过在ORL标准人脸库的实验表明，该算法与现有的方法相比在识别率和识别效率方面都有一定程度的提高，取得了比较理想的效果。

关键词: 二维主成分分析, 双向线性判别分析, 特征提取, 主成分分析, 线性判别分析