Small sample face recognition based on ensemble deep learning

摘要

The problem of small sample in face recognition is still a hot topic nowadays. How to improve the accuracy of face recognition in the case of small sample is very important, for this, this paper proposes a small sample face recognition method based on ensemble deep learning. In order to reduce the problem that the recognition rate decreases greatly due to the too small sample size, first, pixel transformation is carried out on the training database to obtain more training data of different scales; then, the training sets of different scales are classified, and deep learning convolutional neural network (CNN) is used for training respectively; finally, the CNN output results are trained as the meta-features of back propagation (BP) neural network by Stacking strategy of integrated learning, so as to obtain the model data of integrated deep learning and identify it. The ORL face database is used to verify the training sample set of 5 and 6 pieces per person. The experimental results show that the deep neural network using ensemble learning improves 7.5% and 5% respectively compared with the training method using CNN network alone, indicating the excellence of this algorithm.