Convolutional neural networks (CNNs) are wont to achieve unprecedented facial accuracy on large labeled databases. However, smaller organizations wishing to provide secure biometric-based access to its members may find it challenging to train CNNs with minimal computer resources and small training databases. We address this issue by proposing a facial recognition approach and validation based on transfer learning that requires minimal retraining. We show that by adding a single layer of a single trained element with a small number of neurons in a pre-trained network to be able to function and validate authentication can be achieved in small data sets. In addition, new courses can be registered with high monitoring accuracy without properly adjusting the new feature layer.
Article Details
Unique Paper ID: 152422
Publication Volume & Issue: Volume 8, Issue 3
Page(s): 340 - 347
Article Preview & Download
Share This Article
Join our RMS
Conference Alert
NCSEM 2024
National Conference on Sustainable Engineering and Management - 2024