Project information
- Guiding Authority: Tinkeers's Lab, IIT Bombay
- Guide: Satvik Mashkaria
- Project date: 2020
Face Recognition with Liveliness Detection
Facial Recognition systems are the needs of the hours, owing to the fact that biometric sensors are not possible due to current extreme situations. Facial recognition systems can be tricked very easily as it primarily focuses on finding a face and then recognizing it, rather than checking its liveliness. I purpose a face- recognition system which when captured an image would first check for its liveliness and then, further process it for recognition.
What are the various ways one can fool a face-recognizer… Printed photo attack, Printed mask attack, Displayed video attack on mobile phone (iPhone) and Displayed video attack on HD screen (iPad). A Convolutional Neural Network(CNN) can be trained for this which would end up in answering between 0 and 1. 0 means it’s an attack and 1 means it is a live person. So, the image would be first processed via some system which would check the liveliness of the photograph and further go to the recognition step if and only if the outcome of the liveness detection system is greater than some threshold.
The Face Recognition Neural network would be basily a transfer learning model(see the references for the model ). The network would use a Siamese network, where the network uses inception model for training( A Siamese neural network is an artificial neural network that uses the same weights while working in tandem on two different input vectors to compute comparable output vectors.) This model will help one to identify which person's image is.