Abstract: This paper proposes an algorithm based on the l2-norm minimization which is much better than the l1-norm minimization. This algorithm is much more accurate on some database and efficient than the traditional sparse representation-based classification. The algorithm adds a feature dictionary into the training dictionary, which can increase the proportion of the feature information and raise the recognition rate. In a series of experiments, it can be found that the method is more accurate in the recognition rate and robust to both pixel corruption and block occlusion than other methods.

Key words: face recognition, dictionary expansion, l2-norm minimization

摘要： 利用比l1-范数最小化更高效的l2-范数最小化算法，提出了一种在多种人脸数据库上整体更为准确，且比经典基于稀疏表示的人脸分类算法更高效的人脸识别算法。它在传统的训练字典中加入了一个特征矩阵，增大特征信息在字典矩阵中的比重，从而提高识别的准确性。在一系列的实验结果中得出，该人脸识别算法比现有的其他几种典型算法更加准确，而且对噪声和遮挡块的抗干扰性也更强。

关键词: 人脸识别, 字典扩展, l2-范数最小化