Human faces encode a vast amount of information, including distinctive features of an individual and demographic characteristics such as a person's age, gender, and weight. Such information is referred to as soft biometrics, which comprises physical, behavioral, or adhered human characteristics classifiable into predefined human-compliant categories. Additionally, some descriptors, like heart rate, fall into the category of the so-called hidden biometrics, metrics of human physiological activities invisible to the naked eye that can serve to assess a person's health status.
The goal of this thesis is to explore the estimation of biometric traits, namely gender, age, weight, and heart rate from facial visuals. In particular, this manuscript includes contributions on improving deep learning models for automatic and contactless estimation of these traits and seeks to deepen the understanding of the key role that widespread practices of social media uploading and filtering of visuals play in these models and their final prediction.
