Northwind Electronics ENTERPRISE
Act as another user (no password - this is a study lab). Switching user switches company, branding and feature flags.
Northwind Electronics
Siam Sport & Outdoor
Blue Harbor Home
#499

Face Recognition

Enrol people with a few photos, then identify (1:N) or verify (1:1) a new photo or webcam shot.

Technique: Local Binary Pattern Histograms (LBPH) + chi-square distance + threshold
Read the code: src/main/java/com/study/advlab/face/LbphRecognizer.javasrc/main/java/com/study/advlab/face/FaceService.javasrc/main/java/com/study/advlab/face/SyntheticFaces.java
The demo people are drawn by code (SyntheticFaces) - no real photos are used. Face images are biometric data: in a real system you need the person's consent and must protect and delete them properly (PDPA / GDPR). Webcam shots are processed on this PC only.
Who is this?
Recent matches
ModeNearest personDistanceResultWhen
IDENTIFYFah Srisuk0.2445 Match 27 Oct 22:09:31
Enrolled people 6 in Northwind Electronics
  • Anna Lee Sales
    Enrolled face
    Enrolled face
    Enrolled face
    Enrolled face
  • Ben Carter Warehouse
    Enrolled face
    Enrolled face
    Enrolled face
    Enrolled face
  • Chai Sakda Support
    Enrolled face
    Enrolled face
    Enrolled face
    Enrolled face
  • Dara Phan Finance
    Enrolled face
    Enrolled face
    Enrolled face
    Enrolled face
  • Ethan Moore IT
    Enrolled face
    Enrolled face
    Enrolled face
    Enrolled face
  • Fah Srisuk Marketing
    Enrolled face
    Enrolled face
    Enrolled face
    Enrolled face
Threshold = trade-off

A low threshold rejects more strangers (fewer false accepts) but also rejects more real people (more false rejects). Door locks choose low, photo tagging chooses high. Try both with the slider.

Besides the threshold, a match must be clearly closer than the next person (best / second best <= 0.95) - a stranger looks "a bit like everybody".