Deep Learning in Biometric Recognition: Applications and Challenges

Biometric recognition serves as a crucial method for security authentication. With the continuous advancement of artificial intelligence technologies, the application of deep learning (DL) in constructing of biometric recognition systems has become increasingly prevalent. This manuscript delivers a comprehensive overview of the DL techniques used in biometric recognition systems. We systematically analyze the pros and cons of traditional machine learning compared to DL methods in tackling challenges within biometric recognition. Subsequently, the latest DL-based applications in biometric recognition filed and the SOTA DL techniques are summarized. Furthermore, a thorough introduction of adversarial attack technologies, their implications as well as the defense strategies are discussed. Finally, we offer insights into the current state and prospective trends of biometric recognition research.