Deep Learning in Biometrics: A Survey

Abstract

Deep learning; biometrics; fingerprint; ocular recognition; convolutional neural networks; electrocardiogram Deep learning has been established in the last few years as the gold standard for data processing, achieving peak performance in image, text and audio understanding. At the same time, digital security is of utmost importance in this day and age, where everyone could get into our personal devices like cellphones or laptops, where we store our most valuable information. One of the possible ways to prevent this is via advanced and personalized security: biometrics. In this survey, it is considered how the scientific advances in the field of deep learning are applied to biometrics in order to enhance the protection of our data. Firstly, a study will be conducted on tackling ocular identification of twins using deep learning. Then, an improved method for avoiding fingerprint spoofing will be presented, thus solving this method's biggest issue. Finally, a brand new method for biometric identification is proposed based on the usage of the user's electrocardiogram. On every one of these methods, the results manage to top the standard alternative performances.

Roadmap on signal processing for next generation measurement systems

“…SIFT, SURF, HOG, etc), biometric features are often based on intrinsic, trait-specific discriminating properties like face geometry, fingerprint minutiae, or gait patterns. However, with the revolution introducing data-driven and learning-based signal processing, also biometric technology has seen the application of DL [118][119][120][121][122][123], still with some years delay.…”

Deep Learning and Machine Learning, Better Together Than Apart: A Review on Biometrics Mobile Authentication

“…In recent years, biometrics-based authentication methods have shown promising results when tested with machine learning and deep learning algorithms. Deep learning algorithm-centric methods have gained popularity as of late due to recent tech advances enabling their efficiency, such as an increased availability of deep neural network (DNN) training datasets and increased computational power [3].…”