Technology - Face Recognition

Face Recognition

Face recognition, whether performed by a human being or a computer system, is a subtle and complex process. Though it can often take place in the blink of an eye, it nevertheless involves a multiplicity of perceptual, analytic and decision-making tasks. To achieve accurate and robust performance with an automatic facial biometric system, optimal approaches are required to each one of these tasks.

OmniPerception Affinity™ Face Recognition software segments and optimises each stage of the process and provides adaptive algorithms specifically designed to do each job optimally well. A breakdown of the process can be summarised as follows:

1. Image Capture to ISO 19794-5 (complying with the global standard for passport images)
2. Automatic Face Detection (checking whether there is a face in view) 
3. Automatic Localisation (pinning down exactly where the face is in the image) 
4. Manual Localisation tool (optional: for very fine tuning of eye position by hand, if required) 
5. Geometric Normalisation (standardisation of the image to have the face exactly where it should be – for recognition and comparison purposes) 
6. Photometric Normalisation (compensation for any unwanted light effects in the image to present the face with illumination as close to standard as possible) 
7. Facial Image Quality Estimation to ISO 19794-5 (re-check against the standard) 
8. Generation of ISO Full-Frontal Facial Image (standard original image)
9. Generation of ISO Token Image (preparation & presentation of standard cropped image) 
10. Generation of Biometric template – OmniPerception’s ‘FacialPIN™’ (coded facial ID)
11. Verification – 1-to-1 matching (checking that the face presented is the right one) 
12. Identification - 1-to-N (searching in a data base to find a match for the face presented)

Verification (“1-to-1” matching) and Identification (“1-to-Many”, or “1-to-N” matching) are the two main types of facial biometric checking. “Watch List” applications are often required, using data bases of known faces - of VIPs, known undesirables, or other special-case people. These are often referred to as “1-to-Several” matching tasks. They are in fact a sub-category of 1-to-N matching, but with the value of “N” usually less, often much less, than 1,000,000.

Applications requiring large scale 1-to-N face matching, such as in Immigration and Passport Control, can require processing speeds of over 10m faces per second.  OmniPerception’s Colossus™ facial search engine is specifically designed to perform accurately and effectively at this multi-million scale, in both stand-alone and web-based applications.