提出一种基于认知的人脸识别方法:从结构信息和特征信息两方面分别进行加工对人脸进行识别。该方法模拟人类认知过程,提取LDA特征表征人脸结构信息,并就此做出初步识别得到最近邻人脸类别。针对挑选出的相似类别人脸采用2DPCA获取人脸特征信息,并基于模糊隶属度进行精确识别。在ORL和Yale人脸库的实验结果验证了该方法是有效的,并且对各种复杂情况具有一定鲁棒性。
A new face recognition algorithm based on the cognitive theory is proposed,recognizing faces by two independent parts: configural information and featural information.This algorithm simulates the process of human cognition.LDA is performed on face image matrix to form configural information,which is used to select the nearest neighborhood classes.And then,2DPCA is used to extract featural information from the selected faces.Furthermore,according to the featural information,an accurate decision based on fuzzy set is made.Finally,ORL and Yale database is used to test,and the experimental results show the algorithm is valid and insensitive in complex conditions.