Face Recognition System with 2D Anti-Spoofing

Over the past two decades face recognition has gained immense interest and is one of the most interesting research areas. The reason for this could be the need for designing automatic recognition and surveillance systems or a human-computer interface. This involves knowledge and contribution not only from various fields of pattern recognition, machine learning, computer vision, and image processing but also in psychology and neuroscience. There are several challenging factors in face recognition like illumination, scale, expression, and pose which are addressed by several researchers to achieve a good recognition rate, but still, there is no robust technique that is available against uncontrolled factors from the environment, hardware, and software of the system. A facial recognition system is a technology that can recognize a human face by pinpointing and measuring facial features in an image. Face recognition allows access to buildings without the need for a key or even faster transits in airport security. But fraudsters can target this face recognition system like any other privacy technology with spoofing. Such a spoofing attack can be quite severe. Hackers will be able to gain access to secure facilities or buildings and homes resulting in the sabotage of critical and confidential data. In this work, a face recognition system with anti-spoofing features is proposed. ResNet50 neural network architecture is used for training the model for face recognition. For anti-spoofing, eye blink detection for photo attacks, and reflection of light for video attacks are incorporated. A prototype is built that implements these functions in hardware.