Windows 95/98/2000/NT/XP. SDKs can be ported to UNIX and Linux upon request.

Accepts any source of visual signals including photographs, live or
recorded video, and digital video files. Accepts artist rendered images.

Head Finding: 50-300 milliseconds depending on scene complexity.

One-to-one matching: < 1 second.

One-to-many matching: Up to 60 million per minute for vector matching depending on hardware and match parameters used

88 byte vector template / 4.5k intensive template.

Technology can support an unlimited # of records.

Detects moving as well as stationary faces.

Technology works optimally when matching frontal images. Face finding detects faces as long as both eyes are visible, i.e. up to 90 degrees in any direction (up-down, left-right, and tilt) from frontal view. Recognition is invariant with respect to pose up to 45 degrees. From 15 to 35 degrees, there will be a slight loss in matching ability. Beyond 35 degrees, more significant loss of matching may occur.

Performs equally well on all races and both genders. Does not matter if population is homogeneous or heterogeneous in facial appearance.

The algorithm focuses on the inner region of the face and has built-in mechanisms that compensate for natural variability in the face. The result is an engine that is robust with respect to changes in lighting conditions, expression, facial hair and hair style.

Explicitly designed to match faces with or without eyeglasses, as long as the eyes are visible and not occluded by glare.

Does not require special lighting or background. Optimal performance is achieved in diffuse ambient lighting. In addition, performance is best if the subject is not back-lit, but this can be compensated for with gain control on video camera. As a rule, if the image is visible to the human eye, then the system will detect it.

Finds faces in any background, plain or cluttered. Recognition is absolutely independent.

Functions with equal performance on color or gray scale images. Requires 24-bit image depth, with a minimum of 320x240 resolution for desktop verification and 640x480 resolution for surveillance

Can detect faces as small as 20x30 pixels or occupying less than 1% of the total image area. Recognition performance is not significantly affected by low resolution facial images. Optimal recognition occurs at head size of 80x120 pixels.