AntiSpoofing for Facial Recognition Through Separable Convolution
Parag Chaudhari, Nilima Kulkarni, Pranav More · 2021
Facial recognition-based authentication has taken light in recent years due to its convenience and lack of need for specialized hardware such as a fingerprint scanning device. It can be used with the cameras present in our mobile phones. Hence many manufacturers are coming up with facial recognition as an alternative to traditional passwords. Intruders, hackers though weren't stopped by this. This research work discusses the various methods that may be used to fool a traditional facial recognition system and ways to remedy them. This work doesn't aim to replace existing facial recognition systems but to add to it, as an intermediate step that would be performed before the recognition task. We shall tackle this problem as a binary classification problem, build a convolutional neural network for the same and compare it with other results. An open dataset LCC FASD Dataset with around 18000 images is used in this work. We obtained an accuracy of 97 % after training for 30 epochs