Advertisment

That Face is Familiar

author-image
PCQ Bureau
New Update

The human ability to look at a face, memorize it, and recall it when you meet the same person again is entering the domain of the computer. Called facial recognition, this technique enables the computer to use your face as your password, or identify criminals by looking at photographs of a crowd and matching these to an existing database.

Advertisment

Already, two companies have come out with face-recognition software.

The advantages of face recognition over other biometric techniques are that it’s non-intrusive and less expensive to set up. The subject of recognition needn’t click on anything or give his fingerprint, for instance, his photograph can be captured even without his knowledge. It’s less expensive in the sense that databases–of employees, national citizens, criminals, etc–may already exist. The hardware required for a small home or office setup is also not very expensive–all you need are standard video cameras with a resolution of at least 320x240 and a frame rate of at least 3—5 fps, a good video card, and a processor with enough speed. After this, all you need to buy is the software.

Like other biometric techniques, recognizing a face involves taking pictures of that face, extracting its features, creating a template from these features, and comparing this to existing templates in a database. Two techniques–‘eigenfaces’ and ‘local feature analysis’–are primarily used for this.

Advertisment

Eigenfaces

Also called PCA (Principle Component Analysis), this technology is patented at MIT, and is being used by Viisage in its face-recognition software.

Pictures of millions of faces are used to build the eigenface database. These pictures are first compared to see if any features are common between them, and are then used to build the database, which comprises two-dimensional, grayscale, global (taking the face as a single entity) images, each of which has different light and dark areas to highlight different characteristics of a face. The premise is that any face can be built by combining 100—125

eigenfaces.

Advertisment

To begin facial recognition, a photograph of the subject is used to to create a ‘template’, by encoding the face’s characteristics into numbers. This template is compared to the eigenfaces’ database, which throws up results–the faces which the template matches most closely.

The disadvantage of eigenfaces is that since it considers global images, a change in facial expression of the subject–smiling or frowning, or a change in pose or lighting–skews the results. LFA (local feature analysis) does away with this weakness.

Local feature analysis

Advertisment

Currently used by Visionics’ face-recognition software, FaceIt, LFA takes a face’s features to be its building blocks. It bases its results on individual features and their relative distances from each other. So, if one feature changes, as in the case of a smile, features around it will also change, and LFA will be able to consider this change. FaceIt, for example, identifies 80 nodal points–like the distance between eyes, depth of eye sockets, cheekbones, jaw line or chin–that define a face. These nodal points are used to create a faceprint, a numerical code for that particular face. Each faceprint is compressed to 84 bytes in size and stored. This is then compared to faceprints stored in the existing database to come up with matching results.

Other techniques used for facial recognition include neural networks and AFP (Automatic Face Processing) but these haven’t been implemented widely yet.

Face recognition can be used widely, from authenticating people for access to computer networks, bank ATMs, etc, through smart cards, to searching for terrorists or criminals in crowded places (FaceIt was recently used for this on specatators at a football match in Tampa); for surveillance or on voters’ ID cards or passports to prevent fraud. Taking this to the extreme, however, may soon lead to a day when ‘Big Brother is watching you’ becomes an everyday reality.

Pragya Madan

Advertisment