Frontal Face Generation Based Multi-angle Face Identification System

Precise identity recognition is a pre-condition for robots to enter the human living environment. Most of the existed face identification methods cannot work on the non-frontal face since the severe texture loss. In this paper, we propose a novel system to deal with multi-angle face identification in video sequence based on frontal face generation, which replaces the process of detection, alignment in the typical face identification system. To solve the problem of face texture loss in large pose variation, we creatively combine generative adversarial networks (GAN) with the state-of-the-art facial landmark localization method. The proposed system was tested on video database containing multi-angle faces, and the experimental results indicate that our system can recognize more faces in the frames, and improve the accuracy of identification for multi-angle face by 130%.