facedetection

Face detection is a computer vision task that identifies the presence and location of human faces in images or video. The typical output is one or more bounding boxes with optional confidence scores, indicating where faces appear. Face detection is distinct from face recognition, which attempts to identify or verify a person.

Early methods used hand-crafted features such as Haar-like features with AdaBoost in a cascaded classifier (the Viola-Jones framework) to enable real-time detection on modest hardware. Modern approaches rely on deep learning, using convolutional neural networks to directly predict face boxes. Common detectors include general object detectors adapted for faces and specialized networks like MTCNN or RetinaFace.

Typical pipelines include preprocessing, face proposal or detection, non-maximum suppression to combine overlapping boxes, and optional facial landmarks or alignment. Outputs usually provide bounding boxes, confidence scores, and sometimes facial keypoints used for further processing.

Common benchmarks include WIDER FACE, FDDB, and AFW. Evaluation uses metrics such as precision, recall, average precision, and intersection-over-union thresholds. Performance depends on pose, illumination, occlusion, and image quality.

Applications range from photo tagging and video indexing to surveillance, access control, and human–computer interaction. Limitations include bias across demographics, false positives and negatives, and privacy concerns. Responsible use emphasizes consent, data minimization, on-device processing, and regulatory compliance.