Research on Face Recognition Based on Improved Dropout Algorithm

Aiming at the problem of overfitting caused by deep convolutional neural networks in small and medium face databases, based on the idea of Dropout, this paper proposes a method of scaling the activation values of its training and testing phases, which is conducive to defining the model at one time and the training and testing process can be carried out on the same model only by changing its hold/abandon probability. Experiments show that this method can effectively alleviate the problem of overfitting and achieve regularization to a certain extent. At the same time, using a specific face image database for training, after repeated training, the test set accuracy was improved by 1.5%.