This survey provides a comprehensive review of recent advancements in deep face recognition (FR), covering algorithm designs, databases, protocols, and application scenes. The authors highlight the evolution of network architectures and loss functions, categorizing face processing methods into "one-to-many augmentation" and "many-to-one normalization." They also compare commonly used databases for model training and evaluation, and discuss scenario-specific challenges such as cross-faction, heterogenous, multiple-media, and industrial scenes. The survey emphasizes the technical challenges and promising directions in deep FR, including adversarial samples, algorithm/data biases, and model interpretability. The introduction reviews the history of FR, from early holistic approaches to deep learning-based methods, and the significant improvements in accuracy and robustness. The survey then delves into the components of FR systems, including face detection, alignment, and deep feature extraction, and discusses the evolution of network architectures and loss functions. It also explores face processing methods, such as 3D models and GANs, and their applications in training and recognition. Finally, the survey highlights the importance of joint alignment-recognition networks and the use of metric learning and transfer learning in face matching.This survey provides a comprehensive review of recent advancements in deep face recognition (FR), covering algorithm designs, databases, protocols, and application scenes. The authors highlight the evolution of network architectures and loss functions, categorizing face processing methods into "one-to-many augmentation" and "many-to-one normalization." They also compare commonly used databases for model training and evaluation, and discuss scenario-specific challenges such as cross-faction, heterogenous, multiple-media, and industrial scenes. The survey emphasizes the technical challenges and promising directions in deep FR, including adversarial samples, algorithm/data biases, and model interpretability. The introduction reviews the history of FR, from early holistic approaches to deep learning-based methods, and the significant improvements in accuracy and robustness. The survey then delves into the components of FR systems, including face detection, alignment, and deep feature extraction, and discusses the evolution of network architectures and loss functions. It also explores face processing methods, such as 3D models and GANs, and their applications in training and recognition. Finally, the survey highlights the importance of joint alignment-recognition networks and the use of metric learning and transfer learning in face matching.