Reducing the False Alarm Rate for Face Morph Detection by a Morph Pipeline Footprint Detector

In this paper, we introduce a novel multi-level process to reduce the false alarm rate (FAR) of existing state-of-the-art face morph detectors, designed to counter the threat that face morphing attacks represent for face image based authentification scenarios. Therefore, we design a novel morph pipeline footprint detector and a novel verification engine to validate the classification results of these existing detectors. The detectors are based on Benford features derived from JPEG DCT coefficients (in the face region of the image and background) and local derivative pattern features. We evaluate the morph pipeline footprint detector with more than 30,000 images and our morph verification engine with false classified authentic images of state-of-the-art approaches. The evaluation shows that our approach is able to reduce the false alarms by 83.67 %.