Face Spoof Detection by Motion Analysis on the Whole Video Frames
Heni Endah Utami, Hertog Nugroho · 2017
Since the popularity of social media has increased, the spoofing attack is one of the many issues that occur in the face authentication system. Video replay attack is the most vulnerable type of spoofing attack, because distinguishing the face spoofing video from its original is very difficult. Most of the existing studies to detect face spoofing attack focused only on the face area and on static images. These approaches have weaknesses since in the real situation, the background texture and other body parts such as hair and shoulders have also been captured by the camera. In this paper, a face spoofing detection method using motion analysis in the whole video frame was discussed. To optimize the segmentation process between the object and the background area, the model-based segmentation was implemented so that all parts of the object`s/user's body can be correctly separated from the background. The data used are 1200 videos from IDIAP Replay-Attack database and 500 videos from synthesized database for additional experimental data. The results of the experiment were that Half Total Error Rate (HTER) were 4% for IDIAP Replay-Attack data and 1% for synthesized data.