Face recognition is a one of the prominent biometric technique for the identifying or verifying individual. Face recognition technology has a wide range of applications, including authentication, access control, finance, criminal investigation, military, and daily life. In recent years, researchers have implemented various deep learning based approaches for face recognition. Deep learning, particularly deep convolutional neural networks (CNN), has gained prominence in face recognition, with several deep learning models. Transfer learning is a popular deep learning technique that uses a pre-trained neural network model rather than recreating the network from scratch on a new problem. Transfer learning approaches are more accurate, faster, and easier to deploy. In this paperwe applied transfer learning based GoogleNet, and SqueezeNet deep learning architectures for face recognition. We developed our own dataset to train and test the proposed work. Finally, we got an overall accuracy of 89 % on GoogleNet and 93 % on SqueezeNet. The results show that our face recognition model outperforms the most advanced models.

Face recognition, deep learning, Transfer learning, GoogleNet and SqueezeNet