Abstract: The feature selection in face recognition based on rough sets theory’s significance of attribute is proposed.At first，on the basis of PCA method，the feature vectors are extracted and the decision table of rough set is built.Then four definitions，which are classified significance and similar significance for single attribute and attribute subsets，are given respectively.At last，attributes reduction based on classified significance of attribute is proposed，and using similar significance of attribute，the final features for face image recognized classification are selected.The new feature selection method entirely relays on the minor knowledge of the data themselves.So the optimal feature set could be selected，and the face recognition precision could be improved.The experiment results show that the proposed method is superior to the traditional ones.

Key words: face recognition, feature selection, rough set theory, global significance of attribute

摘要： 提出基于粗糙集理论属性全局重要度的特征选择方法改进人脸识别中的特征向量的表征能力。以PCA方法得到的特征向量为基础，给出粗糙集的单个特征和特征子集的属性类间分类重要度和属性类内相似重要度的概念。提出基于属性类间分类重要度的属性约简方法，并用属性类内相似重要度进行最后的特征选择，得到进行人脸图像识别分类器的特征向量。新的特征提取方法完全依赖数据本身的先验知识，可选择出最优的特征组合，提高人脸识别率。实验结果表明，与其他方法相比该方法是有效的。

关键词: 人脸识别, 特征选择, 粗糙集理论, 属性全局重要度