This study is concerned with the reliability of biometric verification systems when used in forensic applications. In particular, when such systems are subjected to targeted impersonation attacks. The authors expand on the existing work in targeted impersonation, focusing on how best to measure the reliability of verification systems in forensic contexts. It identifies two scenarios in which targeted impersonation effects may occur: (i) the forensic investigation of criminal activity involving identity theft; and (ii) implicit targeting as a result of the forensic investigation process. Also, the first partial countermeasure to such attacks is presented. The countermeasure uses client-specific Z-score normalisation to provide a more consistent false acceptance rate across all enrolled subjects. This reduces the effectiveness of targeted impersonation without impairing the systems accuracy under random zero-effort attacks.
ASJC Scopus subject areas
- Signal Processing
- Computer Vision and Pattern Recognition