Biometric recognition is the art of recognising individuals based on intrinsic characteristics, like fingerprints, face video, or iris images. These characteristics are always carried by the person, and are harder to forget, misplace, steal, or copy than ID cards or passwords.

My research is focused on developing more effective, robust, and secure biometric systems. I have focused mainly on the electrocardiogram as a biometric trait, given its hidden nature that makes it more difficult to steal or counterfeit. However, I have also worked on other traits, especially face image and video.

While biometric recognition is the main focus of my research, I have also delved into the topics of template security, presentation attack detection and, more recently, the use of interpretability tools to explain decisions given by deep biometric models.

When looking at a picture, humans immediately recognise there are people in it, maybe some trees, a house, and the sky in the background. For a computer, however, a picture is no more than a matrix of organised numeric values. The same applies to signals, video, or any other data.

In Pattern Recognition, a subarea of Artificial Intelligence, researchers focus on making computers able to detect, discriminate, and identify patterns in data. For that, we use signal processing, image analysis, and/or machine learning methods for computers to learn to understand data.

My research focuses mainly in pattern recognition for biometrics. However, I have also worked on image analysis for automatic medical diagnosis, and on more fundamental machine learning topics including weakly-supervised learning and ordinal classification.

The electrocardiogram (ECG) is the signal that measures the flow of electrical currents across the heart. It is not only a key tool for clinical diagnosis of heart conditions, but a carrier of valuable personal information on identity, health, and emotional states.

Through my research, I create algorithms that use the ECG for automatic recognition of identity, emotions, and fatigue in human-computer interfaces. Specifically, I focus on challenging off-the-person signals acquired unobtrusively, through electrodes seamlessly integrated in common objects such as wearables, smartphones, or steering wheels.

One of my main goals is to make the ECG biometric algorithms as effective and robust as face or fingerprint biometrics, while reataining the higher security it offers as a hidden biometric trait.

The variability of a person's traits is often a nuisance for biometric recognition. However, it often reveals valuable information that could be used to understand other aspects and characteristics of the subjects, including their health and emotional states.

Face image and video has long been used to understand facial expressions and link them to emotional states, despite claims of unreliability. Physiological traits like the ECG include more intrinsic information on health and emotions, that is much harder to fake than facial expressions.

My research in wellbeing monitoring is mainly focused on combining ECG and face video information for more robust emotion, stress, and fatigue recognition. This work is especially focused on human-machine interface contexts, e.g., for real-time monitoring of vehicle drivers.