Face Search and Recognition Methods

Face search produces candidates, not a self-proving identity

A face-search system detects a face, converts its measurable characteristics into a numerical representation, and compares that representation with an indexed gallery. Start with the face recognition search, follow the responsible face search guide, then use the face-search false-positive review before relying on source pages, names, dates, locations, and consistent biographical context.

No visual-search system provides complete coverage. Relevance depends on whether the sought image was available to the search system, successfully processed, and similar enough to pass the retrieval threshold.

A safer review sequence

1. Use a clear, front-facing image with one unobstructed face.
2. Review several candidates instead of assuming the first result is correct.
3. Open each source and check names, dates, organizations, locations, and other photographs.
4. Record uncertainty when the source context is incomplete or contradictory.
5. Never turn a visual candidate into an accusation or adverse decision without independent evidence.

Accuracy changes with operating conditions

Pose, illumination, blur, compression, occlusion, age difference, gallery composition, and the selected threshold all affect outcomes. NIST evaluates one-to-many systems with both false-positive and false-negative identification measures because a single headline accuracy number cannot describe these tradeoffs.