Abstract:
Facial recognition is one of the most extensively researched fields in computer vision and machine learning domain. The field has particularly benefited from the recent advancements in technology leading to production of better cameras and imaging devices. Another reason for the advancement in this field is its applied need in in a wide range of commercial and law enforcement applications.
Even though the current facial recognition algorithms and systems have reached a certain maturity level but there are still critical challenges in deploying these systems in real world applications. A very common problem is the recognition of the faces in images acquired in outdoor environments with varying levels of illumination and different poses. Even though the facial recognition has significantly improved since its inception, but it is still far away from the perception capabilities of a human mind.
Another major problem faced in facial recognition and other biometric identification techniques is their vulnerability to Presentation Attacks (PAs). Face being one of the most easily accessible biometric characteristics is more susceptible to different types of PAs. Common types of facial spoofing attacks include print and replay attacks (2D spoofs) and 3D masks (3D spoofs).
In our research, we have tried to create an algorithm that can be utilized to detect 2D spoofs efficiently. We have tried to treat the spoof detection as an outlier detection problem where the live images belong to a closed set and the spoofs are outliers to that set.
