Idneo Tech Talk: 3D Facial recognition

#IdneoTechTalk by Biel Tura

Facial recognition is lately gaining a lot of popularity

Recognition systems have been embedded into smartphones, banks or door access control. But, is facial recognition a secure biometric identification method? This talk presents face recognition software which seeks to contribute to resolve this security concern.

Nowadays, most of facial recognition software in the market can easily be fooled by a printed picture of a face due to the lack of depth information. To solve this problem, two solutions are proposed: Stereo vision and IR dot projection. The first one generates a 3D depth map with the help of 2 points of view of the scene. The other one projects an IR dot matrix to generate a 3D reconstruction.

The developed software uses a stereo setup and its divided in three different blocks: The frontal face detector detects and aligns a face within the right camera. If a face is successfully aligned though machine learning algorithms, the left camera is triggered to take the left picture. With the right and left frame, a 3D depth map of the face is generated to verify that the scanned person is real.

Analysis of the person’s identity and 3D model are processed through Deep Convolutional Neural Networks. The detected face is embedded into a 128-d dimensional space were similar faces are projected in the same multidimensional place. This stereo setup provides a secure real-time face authentication method for the future facial recognition tasks.

 

Link to our development capabilities:  https://www.idneo.com/#develop