Biases in Fingerprint Recognition Systems: Where Are We At?

Biometric technologies based on fingerprints are widely deployed in systems of national importance to safeguard our society through prevention of identity theft, access control to sensitive facilities (e.g., airports, military bases) and identification of suspected terrorists. Fingerprint technology is subject to biases depending upon the dataset available and the conditions in which the algorithms are created. Common countermeasures to discovered biases consist of exposing machines to more fresh data, feature engineering, algorithm selection, hyper-parameter optimization and retraining the machine to reduce or eliminate the biased outcome. Although gathering data from several random sources may increase the quality of machine learning approaches, systematic research is needed to understand and minimize biases influencing performance and security of biometric systems. The contribution of this paper includes examining biases affecting fingerprint technology from a holistic perspective as well as enumerating the fundamental challenges encountered by current systems operating in real-world applications.