关键词: 人脸识别, 最小化总投影误差, 线性回归分类, 一元回归分类, 一元旋转子空间

Abstract: For the issue that existing regression classification methods in face recognition do not consider total projection error within classes well, a unitary regression classification method based on minimizing Total Projection Error（TPE） is proposed. Characteristics decomposition is used to find unitary rotation matrix after the projection error matrix within class of all the training data is calculated by projection matrix of each class. Then, unitary rotation matrix is used to convert each training image vector to a new vector space, and the specific projection matrix of each class is worked out. Minimum projection error of each class in unitary rotating subspace is used to finish face recognition. The effectiveness and robustness of proposed method has been verified by experiments on the two common face databases FEI and FERET. Experimental results show that proposed method has better recognition accuracy than several other advanced regression classification approaches.

Key words: face recognition, minimizing total projection error, linear regression classification, unitary regression classification, unitary rotate subspace