Facial Recognition Technology

Facial recognition technology is a biometric technology that analyses a face in a digital image or video to verify or identify a person. It converts facial information into a mathematical representation called a facial template, which is compared with previously enrolled templates.

From image to facial template

Identification begins only when the system has a reference database containing facial templates linked to known identities.

• The system uses face detection to locate and crop a face from the image or video frame.
• It aligns the face using landmarks such as the eyes, nose and mouth, reducing differences caused by position or angle.
• An algorithm extracts distinguishing patterns and converts them into a numerical feature vector or template.
• The newly captured image is called the probe, while stored reference images collectively form the gallery.

How matching produces an identity

The algorithm compares the probe template with stored templates and calculates a similarity score. A decision threshold determines whether the similarity is sufficient to report a match.

• In one-to-one verification, the system tests whether the face matches a claimed identity, as in biometric authentication.
• In one-to-many identification, the system searches a gallery and returns the closest match or a ranked list of candidates.
• A reported match is a probabilistic result, not conclusive proof of identity, and may require human or additional verification.

Accuracy and governance concerns

Performance depends on image quality, illumination, pose, facial expression, ageing and occlusion. Error rates may also differ across algorithms and demographic groups.

• A false match incorrectly links a face to another person, while a false non-match fails to recognise the correct person.
• Changing the matching threshold generally involves a trade-off between these two kinds of error.
• Because facial data can enable identification and surveillance, its use engages privacy under Article 21, recognised in Justice K.S. Puttaswamy v. Union of India (2017).
• Public deployment requires lawful authority, proportionality, data security, accountability and meaningful human oversight.