Geometrix How It Works Page

How it works

FaceVision identity systems leverage the most complete and precise 2D and 3D facial data acquisition systems ever produced - the same enrollment stations used in our automated access control systems are used to push the limits of academic forensics research.

The flowchart on this page explains each step of the FaceVision process, and how it changes the paradigm in facial identity verification. For automated identity verification, the entire process executes automatically to the final output. Qualified law enforcement professionals can access any intermediate output to achieve a new level of automated facial forensics capability.

Raw Images: During enrollment or probe, FaceVision sensors acquire one or more pairs of images from closely spaced pairs of cameras. These cameras and their images are like any others you might use for facial recognition or mug shots - except they provide ten times the resolution of standard mug shots, so you can zoom in to examine a freckle. The sensor's integrated illuminator and exposure control loop ensure that every image is identically well-lit regardless of ambient lighting conditions and skin tone. Each image is stored together with sub-pixel accurate calibration information describing the sensor at the time the image was acquired, so any third-party facial imaging tool or application is fully enabled like never before.

Forensic Images: Automated algorithms apply lens calibration information to remove all camera distortions from the images. This improves accuracy of any subsequent image analysis task, and enables immediate use of forensic tools utilizing ratios of distances between facial landmarks visible in the image. To obtain absolute XYZ distances between landmarks at sub-mm precision, the landmark need only be selected in two images.

3D Features: Our patent-pending algorithms find every tiny visual feature on the facial surface in one image and match them to the same features in the other image. Since the cameras are a known distance apart, the algorithms can then triangulate to determine each point's distance from the cameras - at up to a million points on the face! Our identity sensors do this from one view. This output enables a forensic analyst - or our automated recognition engine - to assess a suspect's craniofacial shape to sub-mm precision.

3D Model: Our high-speed algorithms "connect the dots" in 3D to form a surface from the 3D features, and then project the original images onto the surface to give each point on the 3D surface the correct color of the subject's face. This provides a unique capability for exception handling in large-scale ID systems - a guard faced with a questionable ID can call up the 3D model and compare the subject to it from any angle, at any zoom level, under any lighting conditions, using a wide variety of decision making tools.

3D Template: The 3D features are then processed further to extract a biometric template suitable for precise identity verification. Unlike conventional facial recognizers, the FaceVision template contains explicit shape information, not a statistical abstraction. Because a person's 3D facial shape template is not influenced by camera, lighting, or pose like 2D templates are, it represents a nearly ideal way to differentiate between individuals without degradation by confounding factors. The result? Fully automated ID verification almost as good as a forensic analyst can do - because we use the same math!

Secure Data: Whether it is one template on a smart card or a database full of templates, images, and 3D models, you cannot risk unauthorized access to them. So FaceVision systems apply triple-DES encryption to all data immediately as it is generated. Nothing reads the data unless the system allows it.

Smart Card or Database: Depending on the application, our systems store the encrypted template, 3D model, and/or calibrated images in a file or database. For automated ID systems, our standard product line is available in versions for IC smart cards, laser-printed cards, or any ODBC compliant database with communication in SQL and XML/RPC over secure sockets.

Now that you know how the FaceVision platform works, you can see how it transforms the methods used for identifying people by their faces. Humans have been pretty good at that forever - finally, computers are too.