Face Recognition

Project Aim

This project aims to develop a neural network model that accurately predicts if two facial images belong to the same person or not. This has applications in many fields, mainly in security applications with face unlock.

Members

• Pranav Srinath
• Rohit Yatgiri
• Sachin Doddamani
• Ankush
• Priyanka

Work Done

• Used the ATT Database of faces, to train our model to distinguish between pictures of different people. Here, we created identical as well as non identical pairs that are used as an input to the network.
• The model predicts n-dimensional encodings of the two images, which can be visualized as points in a n-dimensional hypothetical space. We strive to minimize the euclidean distance between two images of the same person. Thus, images belonging to the same person have a small distance and those belonging to different people are far apart.
• We assign the threshold as a hyper parameter which is tuned as the model is trained, and this threshold determines the maximum value of distance between two images for them to be classified as belonging to the same person.
• Obtained results of 99.92% test accuracy on this data set which has a threshold of 0.179.

Future Scope

This project can act as a base for many more of this nature which can be applied on to a more diverse dataset. Further, it can be improved to handle real time applications.

Video demo