Age Estimation from Face Images

An age-estimation system analyzes a face image and predicts an age, age band, or related score. It estimates apparent age from image features; it does not read a government record or establish a person’s date of birth. Actual appearance varies with genetics, health, expression, lighting, image quality, and many other factors. NIST’s Face Analysis Technology Evaluation (FATE) Age Estimation and Verification program evaluates submitted algorithms using defined image datasets and publishes performance results, including demographic analyses. Performance is conditional on the algorithm and test population. A result reported on one dataset should not be assumed for another camera, age range, or demographic group. Some applications are low impact, such as aggregate research with appropriate safeguards; using a face estimate to permit a restricted purchase, determine access, or make a legal decision carries a much higher cost of error. In those cases, a face estimate should not replace an accepted verification process or human review. Teams should define the target age concept, consent and retention rules, and acceptable error before collecting faces. Test false acceptance and false rejection around relevant thresholds, break down error by image quality and demographic group, and give people an alternative path. Face images are sensitive personal data in many contexts, and rules vary. Do not make a decision solely from an estimated age when reliable documentation or a safer procedure is available.