DEVELOPMENT OF ACTIVE LIVENESS DETECTION SYSTEM BASED ON DEEP LEARNING ACTIVENESSNET TO OVERCOME FACE SPOOFING

• Main Author: Adi Nur Fauzi, Dhimas
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73930
• Institution: Institut Teknologi Bandung

The facial verification system in the e-government (SPBE) is vulnerable to face spoofing attacks. Face spoofing can be solved through the liveness detection method. One type of face spoofing that will be in this report is the 3D mask attack. In this case, the method used is active liveness detection, which involves a process of detecting liveness by providing a randomized sequence of questions that the user must follow to be categorized as real. If the user is unable to follow the given questions, they will be categorized as fake. The model used for liveness detection is ActivenessNet, which combines three pre-trained models: emotion detection, profile detection, and blink detection. Testing each of these pre-trained models resulted in an accuracy score of 70% on the validation set for the emotion detection model. This score is the highest compared to other models. For the profile detection model, the author obtained a 95% accuracy score on the validation set containing faces facing left and right. As for the blink detection model, it was tested to detect ten blinks in a video and achieved a recall score of 100% under bright and dark conditions when the user was not wearing glasses. When the user wears glasses, the model's performance is decreased, with a recall score of 80% under bright conditions and 60% under dark conditions. The Active Liveness Detection System, designed as the final project product, was then developed as a website.