Face Recognition Using Local PCA Filters

Yida Wang, Shasha Li, Jiani Hu, Weihong Deng

Abstract: We propose an efficient feature extraction architecture based on PCANet. Our method performs far better than many traditional artificial feature extraction methods with the help of standalone filter learning and multiscale local feature combination. Such structure cascaded by both linear layers with convolution filters and non-linear layers in binarization process shows better adaptability in different databases. With the help of parallel computing, training time is much shorter than PCANet and also more fixed compared to convolutional neural network. Experiment in LFW and FERET shows that such a data oriented structure shows good performance both on stability and accuracy in various environments.