29-04-2013, 04:56 PM
An Introduction to Face Recognition Technology
An Introduction.pdf (Size: 113.99 KB / Downloads: 132)
Abstract
Recently face recognition is attracting much attention in the society of network multimedia information access. Areas such as network security, content
indexing and retrieval, and video compression benefits from face recognition technology because "people" are the center of attention in a lot of
video. Network access control via face recognition not only makes hackers virtually impossible to steal one's "password", but also increases the
user-friendliness in human-computer interaction. Indexing and/or retrieving video data based on the appearances of particular persons will be useful
for users such as news reporters, political scientists, and moviegoers. For the applications of videophone and teleconferencing, the assistance of
face recognition also provides a more efficient coding scheme. In this paper, we give an introductory course of this new information processing
technology. The paper shows the readers the generic framework for the face recognition system, and the variants that are frequently encountered by
the face recognizer. Several famous face recognition algorithms, such as eigenfaces and neural networks, will also be explained.
Introduction
In today's networked world, the need to maintain the security
of information or physical property is becoming both increasingly
important and increasingly difficult. From time to time
we hear about the crimes of credit card fraud, computer breakin's
by hackers, or security breaches in a company or government
building. In the year 1998, sophisticated cyber crooks
caused well over US $100 million in losses (Reuters, 1999).
In most of these crimes, the criminals were taking advantage
of a fundamental flaw in the conventional access control systems:
the systems do not grant access by "who we are", but by
"what we have", such as ID cards, keys, passwords, PIN
numbers, or mother's maiden name. None of these means are
really define us. Rather, they merely are means to authenticate
us. It goes without saying that if someone steals, duplicates,
or acquires these identity means, he or she will be able
to access our data or our personal property any time they
want. Recently, technology became available to allow verification
of "true" individual identity. This technology is based
in a field called "biometrics".
Performance Evaluation Metrics
The two standard biometric measures to indicate the identifying
power are False Rejection Rate (FRR) and False Acceptance
Rate (FAR). FRR (Type I Error) and FAR (Type II
Error) are inversely proportional measurements; For example,
if an ID system tunes its threshold value to reject all imposters
(minimizing FAR), it may also improperly reject some authorized
users (maximizing FRR). Therefore, ID system designers
often provide a variable threshold setting for the customers
to strike a balance.
Generic Framework
In most cases, a face recognition algorithm can be divided
into the following functional modules: a face image detector
finds the locations of human faces from a normal picture
against simple or complex background, and a face recognizer
determines who this person is. Both the face detector and the
face recognizer follow the same framework; they both have a
feature extractor that transforms the pixels of the facial image
into a useful vector representation, and a pattern recognizer
that searches the database to find the best match to the
incoming face image. The difference between the two is the
following; in the face detection scenario, the pattern recognizer
categorizes he incoming feature vector to one of the two
image classes: “face” images and “non-face images. In the
face recognition scenario, on the other hand, the recognizer
classifies the feature vector (assuming it is from a “face” image)
as “Smith’s face”, “Jane’s face”, or some other person’s
face that is already registered in the database.
Variations in Facial Images
Face recognition is one of the most difficult problems in the
research area of image recognition. A human face is not only
a 3-D object, it is also a non-rigid body. Moreover, facial images
are often taken under natural environment. That is, the
image background could be very complex and the illumination
condition could be drastic. Figure 2 is an example of an
image with a complex background.
Pattern Recognition
Due to variants such as viewing angles, illumination, facial
expression and so on, the facial feature vector obtained from
previous equations can have random variations and therefore
it is better modeled as a random vector. If the incoming person
is equally likely to be any person in the database (equal a
priori probability), then according to Bayes decision theory,
the minimum recognition error rate can be achieved if the
recognition is following the maximum-likelihood (ML) criterion.
That is, suppose Y = f(X) is the feature vector and suppose
that there are K persons in the database.
Face Recognition Algorithms
In the previous section we have shown that the task of face
recognition encounters complex variations. In order to cope
with such complication and find out the true invariant for recognition,
researchers have developed various recognition algorithms.
In this section, we will describe two representative
ones. The eigenface approach applies the Karhonen-Loeve
(KL) transform for feature extraction. It greatly reduces the
facial feature dimension and yet maintains reasonable discriminating
power. The neural network approach, though
some variants of the algorithm work on feature extraction as
well, mainly provides sophisticated modeling scheme for estimating
likelihood densities in the pattern recognition phase.
Eigenface
As mentioned, one of the goals that the feature extraction routine
wishes to achieve is to increase the efficiency. One simple
way to achieve this goal is using alternative orthonormal
bases other than the natural bases. One such basis is the Karhonen-
Loeve (KL). KL bases are formed by the eigenvectors
of the covariance matrix of the face vector X. In the high dimensional
"face" space, only the first few eigenvalues have
large values. In other words, energy mainly locates in the
subspace constituted by the first few eigenvectors.
Neural Network
In principle, the popular back-propagation neural network
may be trained to recognize face images directly. For even an
image with moderate size, however, the network can be very
complex and therefore difficult to train. For example, if the
image is 128x128 pixels, the number of inputs of the network
would be 16,384. To reduce complexity, neural network is
often applied to the pattern recognition phase rather than to
the feature extraction phase. Sung and Poggio’s face detection
algorithm (Sung, 1995) down-samples a face image into a
19x19 facial feature vector before they apply the elliptical kmean
clustering to model the distributions of the "face samples"
and the "non-face samples". Rowley et al. (Rowley,
1998) also reduce the dimension of the facial image to 20x20
by downsampling before the facial image is fed into their
multi-layer neural network face detector.
Conclusion
Face recognition is a both challenging and important recognition
technique. Among all the biometric techniques, face recognition
approach possesses one great advantage, which is its
user-friendliness (or non-intrusiveness). In this paper, we
have given an introductory survey for the face recognition
technology. We have covered issues such as the generic
framework for face recognition, factors that may affect the
performance of the recognizer, and several state-of-the-art
face recognition algorithms. We hope this paper can provide
the readers a better understanding about face recognition, and
we encourage the readers who are interested in this topic to go
to the references for more detailed study.