TBIOM Special Issue on Trustworthy Biometrics–Editorial
Weihong Deng, Tal Hassner, Xiaoming Liu, Maja Pantić · IEEE Transactions on Biometrics Behavior and Identity Science · 2022
Biometrics are increasingly being used to recognize people in various applications through their physiological or behavioral traits such as face, fingerprint, iris, gait, signature, and voice, and are becoming an integral part of our daily life. A broad and sustainable deployment of biometric systems relies heavily on the ability to trust the recognition process and secure handling of the output. As a result, privacy and security are also critical to the success of biometrics in addition to the high accuracy. To trust a decision made by an algorithm, we need to ensure that it is fair and causes no harm. Building trustworthy systems requires learning unbiased representations so that algorithms are fair to all users (i.e., are handling the imagery of everyone in the same way). Although the accuracy continues to improve, adversarial attacks have created a public concern that biometric systems can be vulnerable. As trust in biometric systems is based on our understanding of how they work, explainability and interpretability enable systems to explain their recognition process and causes of failures. Finally, human experts or users might choose to interact with the system to enhance our trust. Therefore, it is becoming essential to develop diverse approaches towards achieving fairness, robustness, explainability, transparency, and integrate them throughout the entire lifecycle of a biometric application.