This paper addresses the growing importance of remote user verification in internet-based applications, focusing specifically on AI-based counterfeit identification (ID) detection. To address the lack of real-world data, we propose a privacy-preserving patch-based methodology and provide a new public database, FakeIDet2-db, containing over 900,000 real and counterfeit ID patches. Furthermore, we present a novel privacy-preserving counterfeit ID detection method, FakeIDet2, and a reproducible standard benchmark that incorporates existing databases. Patches extracted from 2,000 ID images acquired under various smartphone sensor, lighting, and height conditions are tested against three physical attacks: printing, screen, and synthesis.
