Abstract:Face verification is important for personal identity authentication, which is significant in system security and criminal identification. Face verification task is to give a pair of face images to determine whether they are of the same identity (i.e. binary classification). The traditional authentication method consists of two steps: feature extraction and face verification. In this study, a hybrid convolutional neural network (HBCNN) is proposed for face verification. The main process is divided into three steps: feature extraction, feature selection, and face verification. The key point of this model is to directly use the mixed convolutional neural network to learn the relevant visual features directly from the original pixels and to further process the features through univariate feature selection and principal component analysis (PCA). This can be achieved from the original pixel extraction to a better robustness and expression of the characteristics. The support vector machine (SVM) at the top level is used to see if it is the same person. Experiments show that the mixed convolutional neural network model has a better performance than the traditional method in verifying accuracy of face verification.