Biometric technology facilitates an increasingly important role in modern security, administration, and business systems, including fingerprint, retinal scanning, voice identification, and facial recognition. In essence, face recognition is a broadly applied task of biometric technologies and a quickly advancing task.
The core significance of face recognition is imparting a discreet, non-interrupting means of detection, identification, and verification, without the requirement of the subject-s knowledge or consent. The application of face recognition covers a wide range, such as consumer applications, law enforcement, industry effectiveness, and monitoring systems.
Applicative domains of face recognition technology are automobile security, access control, immigration, education, retail, and healthcare. Some of the hurdles faced by face recognition are posed changes, presence or absence of structuring elements, facial expression changes, aging of the face, several illumination conditions, image resolution and modality, and availability and quality of face dataset. Different machine learning and deep learning approaches have been developed to resolve the abovementioned issues.
Due to the recent advancement of affordable, powerful GPUs and abundant face databases, the research focus of face recognition primarily aims to implement deep learning techniques for all aspects of face recognition tasks. Several deep learning techniques have been proposed to achieve accurate face recognition with constituents such as face processing, deep feature extraction, and face matching. Some popular and often utilized deep learning techniques are Convolutional Neural Networks (CNN), Autoencoder (AE), Generative Adversarial Networks (GAN), Deep Belief Networks (DBN), and Deep Boltzmann Machine (DBM).
• Convolutional Neural Network (CNN) - Convolutional Neural Network (CNN) is the most prominent deep learning technique for face recognition and shows excellent results in image applications. Single CNNs, multi-CNNs, different layouts of multi-CNNs, modifying the way of learning kernels, fusing CNNs with other types of modules, and adopting weakly-supervised or unsupervised learning are the variants and techniques of CNN for face detection.
• Autoencoder (AE) - Autoencoder (AE) and its variants also attained much attention in face recognition. AE-s recent strategy is learning the same latent features between different domains.
• Generative Adversarial Network (GAN) - Generative Adversarial Network (GAN) has increased quickly currently. GAN helps to resolve many face recognition problems, such as pose-invariant face recognition, face synthesis, cross-age face recognition, video-based face recognition, and makeup-invariant face recognition.
• Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), and Deep Boltzmann Machine (DBM) are applied for face classification with their compact representations. Besides these deep learning models, various hybrid deep architectures such as AE+DBM, AE+CNN, GAN+CNN, and RNN-DBM have been exploited for face recognition.
• Cross-Factor Face Recognition - Cross-Pose Face Recognition, Cross-Age Face Recognition, and Makeup Face Recognition are some famous cross-factor face recognition using deep learning.
• Heterogeneous Face Recognition - Heterogeneous face recognition includes various deep learning-based face classification tasks such as NIR-, VIS Face Recognition, Low-Resolution Face Recognition, and Photo-Sketch Face Recognition.
• Multiple media Face Recognition - Set/Template-Based Face Recognition, Video Face Recognition, and Low-Shot Face Recognition are multi- or single media face recognition applications.
• Face Recognition in Industry - 3D Face Recognition, Partial Face Recognition, Face Recognition for Mobile Devices, Face Anti-attack and Debiasing face recognition are industrial applications of face identification with the help of deep learning.
