Abstract: Based on the methods Multi-Directional 2DPCA and Non-Iteration Bilateral projection based 2DPCA, a novel method named MDNIB2DPCA（Multi-Directional Non-Iteration Bilateral projection based 2DPCA） is proposed for image feature extraction. This method can extract features in multiple directions from the image matrices and obtain a higher feature extraction speed than MD2DPCA. The results of contrast experiments on gray image datasets show that the proposed method can obtain a higher accuracy by at least 2 percentage points than the previous methods. Furthermore, on the basis of NIB2DPCA based color image recognition method, an improved color image recognition method is proposed which replaces the NIB2DPCA by MDNIB2DPCA. The results of contrast experiments on color image datasets show that the proposed method can also obtain a higher accuracy by about 1 percentage point.

Key words: color face recognition, Two Dimensional Principal Component Analysis（2DPCA）, Multi-Directional Non-Iteration Bilateral projection based 2DPCA（MDNIB2DPCA）, score level fusion, feature extraction

摘要： 在多方向二维主成分分析法MD2DPCA和无迭代双边二维主成分分析（NIB2DPCA）的基础上，提出了多方向无迭代双边二维主成分分析（MDNIB2DPCA）的特征抽取新方法。该方法可以对图像矩阵在多个方向上进行特征抽取，与MD2DPCA方法相比也提高了特征抽取速度。在灰度人脸图像库上的对比实验表明，所提的方法可以提高灰度图像识别率两个百分点以上;进一步地，在基于NIB2DPCA的彩色图像识别方法的基础上，提出了将所提的MDNIB2DPCA替换NIB2DPCA的彩色图像处理的新方法。在彩色人脸库上的对比实验表明，所提方法的识别正确率也可提高约一个百分点。

关键词: 彩色人脸识别, 二维主成分分析法（2DPCA）, 多方向无迭代双边二维主成分分析（MDNIB2DPCA）, 分数等级融合, 特征抽取