Actually what is mean by 3D face recognition? In normal cases we are extracting face encoding s from a 2D image,right? Is 3D face recognition is used for liveness detection? how its possible?
1. What is typically meant by 3D-face recognition? We are usually extracting the face encoding from 2D-images, right?
Yes. The goal is to reconstruct the three-dimensional shape, as well as the texture of a face from a single or multiple images of that person.
In recent years, "the performance of 2D face recognition algorithms has significantly increased with the use of deep neural networks and the use of large-scale labeled training data." Deep 3D Face Ifentification with the latter still being a notable problem because there are so few of them and they are usually needed to better measure and hence train the deep networks. (The referenced paper gives you a nice overview.) Even with enough data, occlusions, lens distortions or possible variations of expressions, etc. make the problem tough. So many other tricks and augmentation techniques are needed.
2. How can 3D face recognition used for liveness detection?
3D face recognition is used to differentiate 2D spoofs from actual present people. Using multiple time-delayed images one could possibly track if the face moves like a living person as well, or if it is static or stretching weird like a mask, puppet, or moving display.
Nonetheless, often even advanced 2D image-based techniques fall flat in real-world applications because of the used cameras. They might not have enough resolution to recognize if the input is a video of a face on a high-resolution display as explained here.