An Efficient System for Partial Occluded Face Recognition

Face recognition has attracted a lot of interest and made a wide range of applications in the real world. However the technology’s efficiency is limited to well positioned, clear face images, which is not always the case in reality. In this paper, we present a software system that is resilient to various image quality degradation, such as face occlusions, illumination effects and face postures. The system exploits deep learning to build an effective model for face detection, and an Elastic Graph Matching based method to extract the key face features for comparison with the face library. The system also provides flexible functionality and extensions to adopt other biometric recognition algorithm for cross-checking and provides a higher reliability. Experiments show that the proposed system offers a good performance and is robust to face occlusions.