Biometrics refers to measurement and statistical analysis of people’s unique physical and behavioral characteristics for verification and identification purposes. A biometric system is mainly composed of acquisition, feature extraction and comparison modules.

During acquisition, the biometric trait of an individual to be recognized is collected via biometric sensors. The data can be captured under various conditions: controlled or uncontrolled environment, with or without subject knowledge or cooperation, close by or from a distance, indoor or outdoor, etc. These factors can introduce huge variations in the biometric samples which in return can deteriorate the matching performance.

In the feature extraction step, the collected raw data is processed to reduce the data dimensionality while preserving the identity information. Features are expected to be invariant to different acquisition conditions. Finally, the extracted features are compared against the template set in the database either for identification or verification.

Previously, the best features for a biometric recognition task were carefully and mostly manually crafted by researchers; but recently, instead of being engineered, they are automatically learned using deep neural networks. For this reason, there is no more a clear distinction between feature extraction and matching modules, since they are incorporated in a single network.