Face Recognition

Face recognition studies how to verify whether two samples show the same person or identify someone in a gallery despite changes in pose, lighting, age, occlusion, image quality, and sensing modality. Most systems learn discriminative embeddings with convolutional neural networks or vision transformers and margin-based losses, while other approaches use quality-aware matching, synthetic data, image restoration, or cross-spectral and video features. Evaluation spans verification and identification benchmarks such as LFW, IJB-B/C, and RFW, alongside low-resolution, masked, and unconstrained settings; researchers also examine subgroup errors, calibration, attacks, template privacy, and explanation quality. A central tension is achieving reliable recognition without sacrificing fairness, privacy, or security: synthetic data and preprocessing may help but do not always transfer to real conditions, and high aggregate accuracy or fixed thresholds do not guarantee safe decisions.