Identity in the Blood Relation: Unraveling the Complexity of Morph Detection in Kinship Biometrics

Shalvika Srivastav, Poulomi Bhattacharya, Akshay Agarwal, Nalini K. Ratha

Abstract: Face morphing attacks present a significant challenge to biometric authentication systems by allowing multiple individuals to share a single identity. While much existing research has focused on detecting morphs between unrelated individuals, a critical gap remains in understanding morphs generated from genetically related subjects, such as parents and children. These kinship-based morphs closely mimic natural familial resemblance, making them difficult to detect using conventional approaches. This work explores the problem of kinship-based morphing by introducing a large-scale synthetic dataset generated using both Open-CV and latent autoencoder-based techniques across various blending ratios. A hybrid detection framework is proposed that leverages identity features extracted from both FaceNet and ResNet50, aiming to capture nuanced facial inconsistencies that may arise in morphs with substantial familial overlap. The study also considers performance across multiple morphing ratios and investigates generalization under unseen synthetic conditions. Additionally, a face recognition experiment is performed across different morph ratios to reflect how the recognition score varies as the morph ratio varies. This research opens new directions for improving the robustness of morph detection systems in the context of realistic kinship-based attacks.