选择性多本征空间的多级人脸识别方法

针对人脸识别过程中仅靠人脸整体特征识别容易出现误识的问题，以及人脸局部特征的重要性。本着由粗到精的学习原则，设计了选择性多本征空间的多级人脸识别方法（SMEM）。首先对人脸划分为整体、上半部、鼻、眼四个本征区域；然后对各本征建立特征空间并构造BP神经网络人脸识别器；最后，以后验概率为依据，选择性调用各级识别器，直到类内阈值和类间阈值均满足设定值的分类为止。经实验证明，此方法有较高的识别精度。

Selective multiple eigenspaces and multistage face recognition

During face recognition,error recognition is a serious problem if only using the whole face.Considering the importance of partial facial features and the coarse-to-fine learning principles,this paper proposes an algorithm of selective multistage face recognition based on multiple eigenspaces.First,face is divided to four areas,such as the whole,the first half,nose and eyes. Then,eigenspaces and BP neural network classifier for four areas are established.Finally,based on the Maximum A Posteriori Probability(MAP),four level classifiers are selectively referred to recognize on different levels,until both of within-class threshold and between-class threshold achieve the preassigned range.It is proved by experience that this method has a high accuracy.