How does a watchlist-based recognition system decide that an incoming face is a "match"?
It builds a biometric template from the incoming image, compares it to every watchlist entry, computes similarity scores, raises an alert when the threshold is exceeded, and a human operator verifies it.
Image databases take the form of watchlists — structured collections of images and biometric templates the system searches for. The recognition process:
- The incoming image is analysed and a template created.
- The template is compared against all watchlist entries.
- Similarity scores are computed for each entry.
- When a score exceeds the threshold, an alert is generated.
- A human operator verifies the match.
Critical factors: the threshold (balances false positives vs false negatives), the watchlist size (affects performance and accuracy), and image quality (decisive for template quality).
Tip: The threshold is a dial between two failure modes — set it low and you flag innocents (false positives); set it high and you miss real targets (false negatives). There's no setting that eliminates both.