Facial Verification

Facial Verification with Procrustes analysis

Compare two faces and return a geometric, explainable same-or-different verdict: verification only, with no identification against a database.

Why verification needs geometry

Faces in real captures vary in pose, scale, and translation. Procrustes alignment normalizes those differences geometrically before comparison, so the verdict reflects shape similarity rather than incidental pose.

Core capabilities

Face detection and landmarking
Procrustes geometric alignment
Embedding comparison
Configurable match threshold
Similarity and distance metrics
Verification-only (1:1), no identification

Key benefits

  • An explainable geometric basis
  • Pose- and scale-invariant comparison
  • Threshold-controllable verdicts
  • No database matching by design
  • Reproducible, parameterized runs

How it works

01
Step 01

Provide two faces

Supply a reference image and a probe image.

02
Step 02

Detect and landmark

Faces are detected and facial landmarks are extracted.

03
Step 03

Align with Procrustes

The probe is aligned to the reference via translation, scale, and rotation.

04
Step 04

Compare and return

Embeddings are compared to produce a match verdict with similarity and distance.

Who it's for

  • Identity and access teams
  • Onboarding and verification flows
  • Researchers validating face-matching methods

Technical characteristics

  • Procrustes alignment (closed-form 2D rotation)
  • Landmark-to-embedding comparison
  • Configurable verification threshold
  • Quality metrics on every run
  • Pluggable face-detector seam

Frequently asked questions

Coming Soon

Two products, in the works

Multi-Frame Super-Resolution and Facial Verification are in active development. Leave your email and be first to know when they launch.