Addressing the challenge of few-shot learning in CNN-based fingerprint recognition systems

Abstract

Over the past decade, deep learning has transformed mul- timedia processing and, consequently, biometrics, due to its ability to learn meaningful features directly from raw data. The effectiveness of this learning depends largely on the availability of sufficient data. In this study, we explore fingerprint recognition as a deep learning classification task, constrained by the availability of only a few examples per class for prediction