Biometric Template Protection for Neural-Network-Based Face Recognition Systems: A Survey of Methods and Evaluation Techniques

As automated face recognition applications tend towards ubiquity, there is a growing need to secure the sensitive data used within these systems. This paper presents survey of biometric template protection (BTP) methods proposed for securing “templates” (images/features) in neural-network-based The BTP are categorised into two types: Non-NN and NN-learned. use neural network (NN) as feature extractor, but part based on non-NN algorithm applied at either image-level or feature-level. In contrast, NN-learned specifically employ NN learn protected from unprotected image/features. We present examples literature, along with discussion categories’ comparative strengths weaknesses. also investigate techniques evaluate methods, terms three most common criteria: “recognition accuracy”, “irreversibility”, “renewability/unlinkability”. expected, accuracy systems generally evaluated using same (empirical) employed evaluating standard (unprotected) On contrary, irreversibility renewability/unlinkability evaluations found be theoretical assumptions/estimates verbal implications, lack empirical validation practical context. recommend, therefore, greater focus evaluation strategies, provide more concrete insights practice. Additionally, an exploration reproducibility studied works, public availability their implementation code datasets/procedures, suggests that it would currently difficult community faithfully replicate (and thus validate) reported findings. So, we advocate push reproducibility, hope furthering our understanding research field.