Deep Composite Face Image Attacks: Generation, Vulnerability and Detection

Face manipulation attacks have drawn the attention of biometric researchers because their vulnerability to Recognition Systems (FRS). This paper proposes a novel scheme generate Composite Image Attacks (CFIA) based on facial attributes using Generative Adversarial Networks (GANs). Given face images corresponding two unique data subjects, proposed CFIA method will independently segmented attributes, then blend them transparent masks samples. We 526 combinations for each pair contributory subjects. Extensive experiments are carried out our newly generated dataset consisting 1000 identities with 2000 bona fide samples and 526000 samples, thus resulting in an overall 528000 image present sequence benchmark attack potential four different automatic FRS. introduced new metric named Generalized Morphing Attack Potential (G-MAP) FRS effectively. Additional performed representative subset both perceptual quality human observer response. Finally, detection performance is benchmarked three single Detection (MAD) algorithms. The source code together be made publicly available: https://github.com/jagmohaniiit/LatentCompositionCode.