f2 provides three different types of face recognition functionality:

•1:1 for identity verification
•1:N to identify an unknown face
•N:N to find multiple database entries for the same person under different names

1:1 Identity Verification for Access Control
This is used to confirm that the person requesting access to a physical location or online service is who they claim to be. To do this the face of the person presenting an access card or other credential is compared with the face in the database of the person to whom the card was issued. f2 encodes each face by finding around 3000 characteristics in the central area of the face (ignoring beards and other facial hair) and transforms these mathematically into a numeric string. This is the unique identifier of this facial image. The two numeric strings are compared – if they match above a user-defined threshold they are deemed to be of the same person.

In the real world 1:1 benefits from 3-4 different shots of a subject being added to the database record on enrollment. In the entry line a video camera can also take a number of shots and compare them all against all the person’s database images. This will overcome situations where the person is blinking, looking down, wiping their face, etc. If any of them match above the user-defined threshold then the match is accepted. So 1:1 is often really Several to Several.

If the match fails, and the individual won’t reveal their true identity, f2’s 1:N capabilities (see below) can be used to determine if the face is in the database under a different name. This is particularly useful to identify someone who’s stolen an access card, and also to detect buddy-punching, i.e. where an employee who wants time off asks a colleague to enter using their access card. Full details of f2 identity verification are in the attached information sheet.




1:N Identification
This is used to identify an unknown face. The first step is to encode all the faces in the database of faces to match against. f2 will do this automatically, as described above. Any unknown face can then be encoded and matched against the encodings of the faces in the database. By holding the encodings in server memory extremely high search speeds can be achieved – f2 can be configured to search at well in excess of 1m faces per second. It returns a list of thumbnails of the top matches in order of Match%.

Major applications of 1:N include:

•Identifying an unknown face
This is the generic use of 1:N searching, by matching a face against databases of any size such as passports, visas, drivers’ licenses, etc.

•Enabling law enforcement to identify arrestees who won’t give their real name
In many instances a suspect when arrested will refuse to give their name, or give one which is false. As many offenders are repeat offenders, f2 can identify them from their previous arrest record

•Real-time video screening
This works best in controlled locations such as an entry point, escalator, elevator, corridor, doorway, etc. At these locations the camera can be positioned at head-height with the face large in the image area. The person will be looking forward, the lighting can be optimally arranged, and in certain environments there are officers on hand to ask them to remove sunglasses etc, and apprehend them if necessary. The frame capture interval can be set by the user. Matches can be displayed in real time alongside the potentially matching face in the database for easy visual confirmation. This information is also saved for subsequent investigation.

•Identifying terrorist threats before an attack
Face Forensics provides a unique wearable threat detection system which can identify the faces of potential threatening individuals on a watchlist who are near guards, sentries, etc. The solution includes a micro-camera, tablet PC, and high-speed communications to a central control centre.

•Identifying missing children on webpages and seized hard drives
For law enforcement f2 can be pointed at selected websites that are known to contain images of missing & abused children to detect whether any of the faces match those on a watchlist. When a computer is suspected of holding illegal images f2 can match every image, even if cropped or reformatted, against all the images in the database, to identify new images which could have been taken by the individual. f2 will also isolate any faces, compare them against a watchlist of missing children, and display any matches.




N:N Detection of Multiple Records of the Same Person Under Different Names
In many cases the same person may be present in a database more than once under different names. An obvious example is to determine if holders of ID cards such as passports & driver’s licenses who have had the document confiscated have reapplied under a different name. To find these requires every face in the database to be matched against every other, i.e. a N:N search. While f2 can be configured to handle this it does generate a lot of matches – if there are 10,000 faces in a database it would involve 50 million comparisons. While f2 can be configured to handle this quickly, there is a risk of a large number of false positives being generated which would require human verification. The ability to use text filters within f2, for example on gender, can significantly reduce this. f2’s N:N module is not constrained to matching faces within a single database. It can also match faces across different databases.