Overview / Usage

Face Recognition is still an area of AI which can be optimized. In this project specifically we try to propose a Deep Learning Model which can output better Face Recognition for Indian Face Images.

Facial recognition is a way of recognizing a human face through technology. A facial recognition system uses biometrics to map facial features from a photograph or video. It compares the information with a database of known faces to find a match.

This can be further used as Automated Face based attendance system in companies, educational institutions, industries etc. This can also be used to manufacture a surveillance solution for many industries.

Methodology / Approach

Many face recognition solutions like FaceNet/Openfaces etc are deep neural network trained to generate 128 facial encodings for input image. The face distance is calculated for 2 faces by comparing these 128 encodings of two facial images. Many facenet models are trained by using datasets like 'Labeled Faces in the Wild', CASIA-WebFace dataset etc, which contains very less or no Indian faces. The accuracy on Indian faces is hence comparatively less for many of these models.We are thus using Indian Movie Face Database for training purposes.

We use Inception-Resnet model because it introduces residual connections that add the output of the convolution operation of the inception module, to the input. .

This would be further optimized using Intel's OpenVINO Toolkit and can be implemented realtime using Intel's Movidius NCS outside a classroom for automated attendance.

Technologies Used

Tensorflow

Intel's OpenVino ToolKit

Intel AI DevCloud

Other links

• Dataset Link
• FaceNet

• FaceNet Github (https://github.com/davidsandberg/facenet)