04 - Different traits recognition


  • Description:: how traits are recognized

Kinds of traits

  • voice: Gaussian
  • face: 3D model, hybrid, feature based (grid)…
  • Iris: eye image capture, unwrapping retina, create a kind of barcode
  • fingerprint: simplification of the fingerprint by applying images filter

Pros and cons


  • cannot be lost, stolen or forgotten
  • the user must only appear in person


  • no 100% accuracy
  • if damaged biometric trait is hard
  • if a trait is copied the user cannot change it
  • unreliable in certain conditions


  • expression recognition is hard to recognize
  • even a single profile or another reflex on the glasses could represent a problem, because the geometry problem changes
  • father and son or twins are hard to spot, even for biometric systems
  • poor quality acquisition
  • bad lighting (not uniform)


  • fake biometric get sensor data
  • replay old data
  • override feature extractor
  • a bad person could modify template, override a final decision and access to information it shouldn’t


A subject is accepted if the similarity achieved from matching with the gallery template corresponding to the claimed identity, is greater than or equal to the acceptance threshold. (or if the distance with such gallery template is less than or equal to the acceptance threshold) Otherwise it is rejected.

Four cases:

  • 📗 Genuine match
  • 🧧False rejection (FNM / Type 1 error)
  • 📗 Genuine reject (impostor is rejected)
  • 🧧 False match (FM / Type 2 error)