Research And Implementation Of Small Sample Face Recognition Algorithm Based On Deep Learning

Face recognition technology has been one of the most important and hottest topics in the field of computer vision,with a wide range of application scenarios.However,when real face images are captured,the number of captured images,the pose of the subject,facial expressions,lighting changes and other factors limit the performance of face recognition algorithms.Deep learning face recognition algorithms need a large amount of training data and deep network structure as support to obtain better results,so there are problems such as overfitting,long training time and complicated calculation.The main objective of this paper is to simplify the training process of deep learning methods for small-sample face recognition and improve its accuracy and robustness on small-sample face datasets.The content and results of the research work in this paper are as follows:(1)A face recognition algorithm based on PCANet and feature fusion is designed and implemented.A multi-feature cascade network model is designed based on the PCANet network model: MF＿PCANet.the features of face images extracted from multiple PCA layers in PCANet are cascaded,thus obtaining more feature information of face images,and the model is used to compare with the traditional face recognition method PCA,the deep learning face recognition method PCANet,the improved PCANet for comparative experimental analysis,and proved that MF＿PCANet is more robust and has higher face recognition accuracy on small sample face datasets ORL and AR.(2)A face recognition algorithm based on the combination of migration learning and attention mechanism is designed and implemented.The idea of transfer learning is introduced,and the channel attention ECANet module is added to the pre-trained network model Res Net50 in the Image Net dataset to construct the Res Net50＿ECANet model,and adaptive histogram equalization with random rotation,cropping,horizontal inversion and restricted contrast is used for face image enhancement;to reduce the training time of the network model,the parameters and weights of the pre-trained model are migrated and its structure is fine-tuned and applied to the target small-sample face dataset.Experiments show that the method proposed in this paper takes 2.2%～47.3% less time than other deep learning network models while guaranteeing face recognition accuracy improvement of 1.71%～5.86%.To verify the effectiveness of the proposed method in this paper,experimental analysis is performed on a small sample face dataset.In the experiments,the proposed method in this paper is compared with the original benchmark method and other deep learning network models.The results show that the method in this paper has high accuracy and performs better on small-sample face dataset.Finally,a face recognition system is designed and implemented to perform face recognition function in real time.The small-sample face recognition algorithm designed in this paper is of reference value for software that requires high real-time and accuracy rate for small-sample face recognition.