基于正交邻域保持投射的DT-CWT特征人脸识别

Face Recognition via DT-CWT Feature-Based Orthogonal Neighborhood Preserving Projections

为了有效地利用双树-复小波变换(DT-CWT)进行人脸识别,提出一种将DT-CWT与正交邻域保持投射(ONPP)相结合的方法.先通过DT-CWT得到具有空间、频率以及方向特征的人脸特征表示,然后使用ONPP对特征向量进行线性降维,有效地保持了数据的局部与全局的几何特征,最后进行识别测试.实验结果表明,基于DT-CWT的ONPP算法可以在特征维数有效降低的前提下很好地完成人脸识别任务.

In order to efficiently accomplish the face recognition using dual-tree complex-wavelet transform（DT-CWT）,a method composed of DT-CWT and orthogonal neighborhood preserving projections（ONPP） is proposed.In this method,a feature vector with desirable characteristics including spatial locality,frequency and orientation selectivity is obtained by DT-CWT.Then,the dimension of the vector is linearly reduced by ONPP,thus effectively preserving the intrinsic neighborhood geometry of the vector and the global geometry.So,the recognition can be easily performed.Experimental results demonstrate that the proposed ONPP based on DT-CWT can accomplish the face recognition well with the feature dimension being effectively reduced.