The Interplay of AI and Biometrics: Challenges and Opportunities (original) (raw)

B iometric applications serve to identify individuals based on their biological characteristics or behaviors. Currently, fingerprints and faces are the predominantly used biometric modalities, but systems based on hand veins, irises, or voices are also available. Biometric applications can use single modalities or combine multiple ones, and they can process static data (for example, for facial recognition) as well as data sequences (for example, for video identification or speaker verification). Data can be acquired using single or multiple sensors of different types (for example, optical or acoustic). Nowadays, biometrics are being increasingly used in many applications in different sectors. Such applications range from automatic border control and physical access control in some contexts to a plethora of use cases in which biometrics are used for authenticating individuals (for example, for authorizing user actions on mobile devices in the consumer sector). This widespread use is mostly based on substantial performance improvements due to the use of connectionist artificial intelligence (AI) methods, in particular, deep neural networks (DNNs). One striking example of their superiority over traditional systems is their ability to match facial images taken from different angles with high probability. 1