Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Face Detection
Algorithmic process that automatically locates human faces in digital images or video streams, typically by delineating rectangles around the detected faces.
Face Recognition
Biometric technology that identifies or verifies a person's identity by analyzing and comparing captured facial features with those stored in a database.
Face Verification
1:1 comparison process that determines whether two facial images belong to the same person, typically returning a binary response (match/non-match).
Face Identification
1:N recognition process that identifies a person by comparing their face with a database of known faces to find a match.
Face Embedding
Low-dimensional vector representation (typically 128-512 dimensions) that uniquely encodes the discriminative features of a face to facilitate comparison.
Landmark Detection
Precise localization of key anatomical points on the face such as eye corners, nose tip, and mouth contours to normalize and analyze facial expressions.
Face Alignment
Geometric transformation process that normalizes the orientation and position of a face using facial landmarks to reduce pose variations.
Liveness Detection
Set of techniques that determine whether a presented face belongs to a living person rather than a photo, video, or 3D mask to prevent spoofing attacks.
Triplet Loss
Loss function used in metric learning that minimizes the distance between an anchor and a positive face while maximizing the distance with a negative face.
ArcFace
Advanced facial recognition architecture that adds an angular margin to the softmax loss function to improve inter-class separation and intra-class compactness.
Cosine Similarity
Similarity measure between two facial encoding vectors calculated as the cosine of the angle between them, ranging from -1 (completely opposite) to 1 (identical).
FaceNet
Convolutional neural network architecture developed by Google that directly learns facial encoding mappings to a compact Euclidean space where distance corresponds to facial similarity.
Anti-Spoofing
Security mechanisms designed to detect and prevent attempts to deceive facial recognition systems using artifacts like photos, videos, or masks.
Face Clustering
Unsupervised process that automatically groups similar facial images without predefined labels to organize large photo collections.
DeepFace
Deep facial recognition architecture developed by Facebook using nine neural network layers and achieving near-human accuracy on the LFW benchmark.
Face Normalization
Set of preprocessing steps including alignment, lighting correction, and dimension standardization to reduce non-identity variations in facial images.
Siamese Network
Neural network architecture with two identical branches sharing weights, trained to determine if two inputs (faces) are similar or different.
Face Tracking
Process that continuously follows the movement of one or more faces in a video sequence while maintaining consistent identification frame by frame.
Attribute Recognition
Facial analysis that predicts demographic or physical attributes such as age, gender, ethnicity, or emotional expressions from a facial image.