With the development of information technology and cyber problems, the technological world focuses on innovative techniques for protecting digital identities. The technical advancement led to the development of extremely sophisticated techniques and authentication technologies, such as biometrics which performs the validation and identification of an individual based on behavioral and physiological attributes.

Biometric features are fingerprints, palm prints, facial features, ears, irises, retinas, signatures, gaits, keystrokes, and voice. Some common applications of biometrics are cell phone authentication, border control, airport security, the automotive industry, screen navigation, and forensic science.

In recent years, deep learning technology has increasingly been supported to enhance the accuracy of several biometric recognition systems. Extensive research contributions have been made in deep learning-based biometrics recognition, as the deep learning models effectively automate the recognition of biometrics.

Learning architectures utilized for biometrics recognition systems are Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short Term Memory (LSTM), Deep Belief Networks (DBN), Auto-Encoders, Generative Adversarial Networks (GANs), Transformers, Capsule Network, Generative Recurrent Unit (GRU), and spatial transformer networks.

Most of the deep learning-based biometrics recognition applies a transfer learning approach. The most promising applications of various biometric recognition tasks using deep learning;

•  Face recognition - Deep learning-based face recognition provides accurate identification and reinforcement learning; multi-task CNN, CNN, and Siamese Network are the currently employed deep learning algorithms. Recent face recognition approaches using deep learning include multiview face recognition, Thermal face recognition, 2D and 3D face recognition, and Age invariant face recognition.

•  Ear Recognition - Ear recognition using deep learning models focusing on small datasets problem and CNN, VGG-16, AlexNet, Faster-RCNN, and pre-trained models are commonly utilized deep learning frameworks.

•  ECG Recognition - Multi-Layer Perceptron(MLP) and Deep auto-encoders are deep learning techniques applied for ECG recognition with the heartbeat and morphological features representations.

•  Finger-vein Recognition - Different variants of CNN, such as ten-layered CNN and LeNet-5 CNN, are used for finger-vein recognition under various imaging qualities.

•  Fingerprint Recognition - Deep learning architectures utilized for fingerprint recognition are ResNet50, RNN, CNN, Artificial Neural Network, and VGG network. Recent fingerprint recognition methods include fingerprint pore extraction for recognition, scalable fingerprint recognition and fingerprint recognition with a 1D representation.

•  Gait Recognition - Deep learning architectures such as DeepCNN, Capsule network, CNN ensemble, Pretrained Densenet, VGG-19, and AlexNet are employed for gait recognition.

•  Signature Recognition - Deep learning architectures such as LSTM, RNN, and CNN are recently applied for online and offline signature recognition.

•  Iris Recognition - Inception V2, Modified CNN, Soft-max classifier, VGG, GoogleNet, ResNet, and AlexNet are the deep learning architectures exploited for iris-based identification. Noise invariant iris recognition has a high focus of interest using deep learning.