Touchscreen gestures as images. A transfer learning approach for soft biometric traits recognition

Mobile devices such as smartphones and tablets are nowadays daily employed by more than 3 billion people, with an expected further worldwide penetration up to 5 billion users by 2025.1 Among the reasons for such astonishing growth, from the early years of mobile communications to the present day, there is the fact that such devices offer the chance to perform different tasks and access several services, such as taking pictures, socializing, finding road routes, or perform online payments with an extreme easiness. As a matter of fact, use of apps like Apple Pay or Google Pay in North America is believed to double between 2020 and 2025, although Asia’s market size will be significantly larger.2 Mobile devices are equipped with a variety of sensors, each designed for capturing specific signals like those related to the heart rate, or more basic ones related to touch gestures. Thanks to these abilities a vast number of applications is being developed for mobile platforms, ranging from activity tracker (Nweke et al., 2018) and healthcare (Shabut et al., 2018) to social recommendation (Gao et al., 2021). It has yet to be observed that most of the services which can be performed through mobile devices are typically accessed and used by providing sensitive and valuable data, such as passwords, credit card numbers, and so on. In this regard, resorting to biometric recognition systems seems a natural choice. Mobile devices’ sensors can be exploited to acquire discriminating traits, thus allowing to recognize the authorized users. Furthermore, the possibility of performing biometric recognition within mobile devices may come in handy to use them as authenticating tokens, providing the means to perform decentralized access control, thus exploiting mobile technology as authenticating means by combining their capabilities with biometric solutions. Within this research area, we also find soft biometrics, an active field of research that provides useful attributes to assist more complex ecosystems. It can improve the performance of biometric authentication systems, user experience in healthcare systems and smart spaces, and play a key role in access control systems. The results reported in the literature (Chai et al., 2019, Idrus et al., 2015, Jain et al., 2004, Park and Jain, 2010, Ranjan et al., 2017) indicate that the authentication performance can be improved by augmenting traditional biometric traits with soft biometric traits, especially when using gender and age.