Abstract
Convolutional Neural Networks (CNNs) have been developed as powerful models for image recognition problems requiring large-scale labeled training data. However, estimating millions parameters of deep CNNs requires a huge amount of labeled samples, restricting CNNs being applied to problems with limited training data. To address this problem, a two-phase method combining data augmentation and CNN transfer learning i.e., fine-tuning pre-trained CNN models are studied herein. In this paper, we focus on the case of a single sample face recognition problem, intending to develop a real-time visual-based presence application. In this context, five well-known pre-trained CNNs were evaluated. The experimental results prove that DenseNet121 is the best model for dealing with practice problems (up to 99% top-l accuracy) is the best and most robust model for dealing with the single sample per person problem, which are related to using deep CNNs on a small dataset and specifically to single sample per person face recognition task.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call