Improved PCA-BP Face Recognition Based on Construction of Virtual Sample

According to the lack of training samples and the shortage of the traditional PCA and BP algorithm, this paper proposes an improved PCA-BP face recognition algorithm based on the construction of virtual samples. Firstly, the algorithm uses original training samples to generate virtual samples by mirror and rotation transformation, and then all of the original samples and virtual samples are used for the improved PCA algorithm to extract the feature vectors. Moreover, the extracted feature vectors are input to the improved BP neural network for training, and finally combined K-Nearest Neighbor algorithm comprehensive discriminant classification, the generalization ability of face image recognition is enhanced. The experimental result on ORL, YALE, FERET and AR face database proves that the proposed algorithm can predict the possible changes of the samples in a certain extent, and the recognition rate is improved obviously.