Robotics Reference
In-Depth Information
make a picture complete. Modern vision systems are therefore usually
only part of a larger system in which other AI technologies play their
part.
Human Face Recognition
Your face is your most unique physical aspect, noticed by anyone who
recognizes you. Different people may have different heights, weights
and shapes, but there are millions of tall and millions of short people
in the world, millions who are fat and millions who are thin. So faces
work much better as distinguishing aspects than bodily shape or size.
One of the most important uses nowadays of computer vision systems
is face recognition, automatically picking someone's face out of a crowd,
extracting that face from the rest of the scene, measuring the various
features of that face and comparing the measurements with those of the
faces in a database of stored images.
A powerful face detection system, developed at Carnegie Mellon Uni-
versity, examines regions of an image in blocks of 20 pixels
×
20 pixels.
×
In order to detect faces anywhere in the original image the 20
20 block
is applied all over the image. The output from this stage of the process is
a collection of these 20
×
20 “windows”, each of which is accompanied
by an estimate ranging from
−
1(defi-
nitely a face present), signifying the system's opinion as to whether or not
a face is there. Each window in this collection is then passed through a
neural network recognition system that examines smaller blocks of pixels
in order to attempt to recognize specific facial features. There are two
smaller blocks designed to detect features such as a single eye, or a nose,
or the corner of a mouth, and there are six stripe-shaped blocks that help
the system to detect features such as a mouth or a pair of eyes.
Having confirmed the presence of a face in an image, the system then
measures its features. Everyone's face has certain distinguishable features,
for example peaks and troughs. There are about 80 of these features on
a human face, including the distance between the eyes, the width of the
nose and the depth of the eye sockets. After measuring these features the
combination of measurements is used by the system to create an almost
unique numerical code that represents a particular face. This code is
called a faceprint. Usually between 14 and 22 of the 80 features in a
faceprint will be sufficient to complete the recognition process. In the
1 (definitely no face present) to
+
