Li S.Z., Jain A.K. (eds.) Encyclopedia of Biometrics
Подождите немного. Документ загружается.
Face Databases and Evaluation. Figure 2 (Continued)
300
F
Face Databases and Evaluation
Face Databases and Evaluation. Figure 2 Examples of performance evaluation conducted on face databases:
(a) identification performance of several appearance-based recognition algorithms measured using CMC curves on
FERET database (from [25]), (b–e) verification and identification performance of commercial face recognition biometrics
systems on FRVT datasets (from [24, 26], using CMC curves (b), ROC curve (c), DET curve (d) and fixed-FAR FRR
distrubutions (e)).
Face Databases and Evaluation
F
301
F
Example Protocols
Feret protocol [7] is an example of the close set
face identification, where a full distance matrix that
measures the similarity between each query image and
each database image is computed. FRVT2002 [24]
addresses both open set verification problem and
close-set identification problem and uses CMC and
ROC to compare the results. BANCA protocol [9],
which is designed for multi-modal databases, is an
example of the open set verification protocol.
XM2VTS Lausanne protocol [13] is an example of a
close set verification, where anyone not in the database
is considered an imposter.
Evaluation Results
Face Databases have been used over the years to com-
pare and improve the existing face recognition techni-
ques. Some of the obtained evaluation results are
shown in Fig. 2. Figure 2a shows face identification
results from [25] for popular appearance-based face-
recognition techniques: Principal Component Analysis
(PCA), Independent Component Analysis (ICA), and
Linear Discriminant Analysis (LDA), obtained on
FERET database using CMC curves.
Figures 2b–e show performance evaluation of com-
mercial FRBSs that participated in the FRVT2002 and
FRVT2006 tests taken from [24 , 26]
Future Work
Considerable advances have been made recently in the
area of automated face recognition. FRBSs are now able
to recognize faces in documents with the performance
that matches or exceeds the human recognition perfor-
mance. In large part, this has become possible due to
the help of many researchers that have collected and
maintained face databases. At the same time, despite
the intensive use of these databases, no FRBS has been
developed so far that can recognize faces from video
with performance close to that of humans.
Automate d recognition of faces from video is con-
siderably worse than face recognition from documents,
whereas for humans it is known to be the opposite. This
status-quo situation serves as an indication that new
evaluation datasets and benchmarks are needed for
testing video-based face recognition system s. Knowing
how easily available have become recently amounts of
various video data (including news casts, televised
shows, motion pictures, etc), it is foreseen that instead
of using video-based data-bases, which are very costly
and time consuming to create, the research community
will soon adopt face evaluation benchmarks and pro-
tocols based on public domain video recordings [5].
The importance of improving the performance of
video-base d face recognition should not be underesti-
mated, taking into account that of all hard biometric
modalities, video-based face recognition is the most
collectable and acceptable [27].
Related Entries
▶ Face Detection
▶ Face Recognition
▶ Identification
▶ Verification
References
1. Gorodnichy, D.O.: Facial recognition in video. In: Proceedings
of the International Conference on Audio- and Video-Based
Biometric Person Authentication (AVBPA’03), LNCS 2688,
pp. 505–514, Guildford, UK, online at http://iit-iti.nrc-cnrc.gc.
ca/iit-publications-iti/docs/NRC-47150.pdf. (2003)
2. Gross, R.: Face Databases. Springer, New York (2005)
3. Face Recognition website. http://www.face-rec.org
4. Face Recognition Vendor Test website. http://www.frvt.org
5. Gorodnichy, D.O.: Seeing faces in video by computers (Editorial).
Image and Video Computing, Special Issue on Face Processing
in Video Sequences (online at http://iit-iti.nrc-cnrc.gc.ca/iit-
publications-iti/docs/NRC-48295.pdf) 24, 1–6 (2006)
6. AT&T.: The Database of Faces (formerly The ORL Database
of Faces, http://www.cl.cam.ac.uk/research/dtg/attarchive/facesa
taglance.html
7. Phillips, P.J., Moon, H., Rizvi, S., Rauss, P.J.: The FERET evalua-
tion methodology for face-recognition algorithms. IEEE Trans.
Pattern Anal. Mach. Intell. 22(10), 1090–1104 (2000). http://
www.nist.gov/humanid/feret/
8. Georghiades, A., Kriegman, D., Belhumeur, P.: From few to
many: generative models for recognition under variable pose
and illumination. IEEE Trans. Pattern Anal. Mach. Intell.
23(6), 643–660 (2001)
9. Bailly-Bailliere, E., Bengio, S., Bimbot, F., Hamouz, M.,
Kittler, J., Mariethoz, J., Matas, J., Messer, K., Popovici, V.,
Poree, F., Ruiz, B., Thiran, J.-P.: The BANCA database and
evaluation protocol. In Audio- and Video-Based Biometric
Person Authentication (AVBPA), pp. 625–638 (2003)
302
F
Face Databases and Evaluation
10. Phillips, P.J.: Human identification technical challenges. In IEEE
International Conference on Image Processing, vol. 1, pp. 22–25
(2002)
11. OToole, A., Harms, J., Snow, S., Hurst, D.R. Pappas, M.,
Abdi, H.: A video database of moving faces and people,
submitted (2003)
12. Martinez, A.M., Benavente, R.: The AR face database. Technical
Report 24, Computer Vision Center(CVC) Technical Report,
Barcelona (1998)
13. Messer, K., Matas, J., Kittler, J., Luettin, J., Maitre, G.:
XM2VTSDB: The extended M2VTS database. In: Second Inter-
national Conference on Audio and Video-based Biometric Per-
son Authentication (1999)
14. Sim, T., Baker, S., Bsat, M.: The CMU pose, illumination,
and expression database. IEEE Trans. Pattern Anal. Mach. Intell.
25(12), 1615–1618, http://www.ri.cmu.edu/projects/project 418.
html) (2003)
15. Hwang, B.-W., Byun, H., Roh, M.-C., Lee, S.-W.: Performance
evaluation of face recognition algorithms on the asian face
database, KFDB. In Audio- and Video-Based Biometric Person
Authentication (AVBPA), pp. 557–565 (2003)
16. Gao, W., Cao, B., Shan, S., Zhou, D., Zhang, X. Zhao, D.:
CAS-PEAL large-scale Chinese face database and evaluation
protocols. Technical Report JDL-TR-04-FR-001, Joint Research
and Development Laboratory, http://www.jdl.ac.cn/peal (2004)
17. Gorodnichy, D.O.: Video-based framework for face recognition in
video. In: Second International Workshop on Face Processing
in Video (FPiV’05). Proceedings of Second Canadian Conference
on Computer and Robot Vision (CRV’05), pp. 330–338. Victoria,
BC, Canada, ISBN 0-7695-2319-6, online at http://iit-iti.nrc-
cnrc.gc.ca/iit-publications-iti/docs/NRC-48216.pdf. (2005)
18. Gorodnichy, D.O.: Recognizing faces in video requires
approaches different from those developed for face recognition
in photographs. In: Proceedings of NATO IST - 044 Workshop
on Enhancing Information Systems Security through Bio-
metrics. Ottawa, ON, Canada, October 18–20 (online at
http://iit-iti.nrc-cnrc.gc.ca/iit-publications-iti/docs/NRC-
47149.pdf. (2004)
19. Shakhnarovich, G., Viola, P.A., Moghaddam, B.: A unified
learning framework for realtime face detection and classifica-
tion. In: International Conference on Automatic Face and
Gesture Recognition, pp. 10–15, USA (2002)
20. Gorodnichy, D.O.: Introduction to the First IEEE Workshop on
Face Processing in Video. In: First IEEE CVPR Workshop on Face
Processing in Video (FPIV’04), Washington DC, USA, online at
http://www.visioninterface.net/fpiv04/preface.html (2004)
21. Willing, R.: Airport anti-terror systems flub tests face-
recognition technology fails to flag suspects, in USA TODAY.
Accessed Sept 4, 2003. http://www.usatoday.com/usatonline/
20030902/5460651s.htm.
22. Yip, A., Sinha, P.: Role of color in face recognition. MIT tech
report (ai.mit.com) AIM-2001-035 CBCL-212 (2001)
23. Wayman, J.L., Jain, A.K., Maltoni, D., Maio, D. (eds.): Biometric
Systems: Technology, Design and Performance Evaluation.
Springer, New York (2005)
24. Phillips, P.J., Grother, P., Ross, J.M., Blackburn, D., Tabassi, E.,
Bone, M.: Face recognition vendor test 2002: evaluation report
(March 2003)
25. Delac, K., Grgic, M., Grgic, S.: Independent Comparative Study
of PCA, ICA, and LDA on the FERET Data Set. Int. J. Imaging
Syst. Technol. 15(5), 252–260 (2006)
26. Overview of the Face Recognition Grand Challenge - IEEE
Conference on Computer Vision and Pattern Recognition,
June 2005. Online at http://www.frvt.org/FRGC
27. Jain, A.K., Ross, A., Prabhakar, S.: An Introduction to Biometric
Recognition. IEEE Trans. Circ. Syst. Video Technol. Special Issue
on Image- and Video-Based Biometrics 14, 4–20 (2004)
Face Detection
MIN G-HSUA N YANG
University of California, Merced, CA, USA
Synonym
Face Localization
Definition
Face detection is concerned with finding whether there
are any faces in a given image (usually in gray scale)
and, if present, return the image location and content
of each face. This is the first step of any fully automatic
system that analyzes the information contained in faces
(e.g., identity, gender, expression, age, race, and pose).
While earlier work dealt mainly with upright frontal
faces, several systems have been developed that are able
to detect faces fairly accurately with in-plane or out-of-
plane rotations in real time. Although a face detection
module is typically designed to deal with single images,
its performance can be further improved if video
stream is available.
Introduction
The advances of computing technology has facilitated
the development of real-time vision modules that inter-
act with humans in recent years. Examples abound,
particularly in biometrics and human computer inter-
action as the information contained in faces needs to be
analyzed for systems to react accordingly. For biometric
systems that use faces as nonintrusive input modules,
Face Detection
F
303
F
it is imperative to locate faces in a scene before any
recognition algorithm can be applied. An intelligent
vision-based user interface should be able to tell the
attention focus of the user (i.e., where the user is looking
at) in order to respond accordingly. To detect facial
features accurately for applications such as digital cos-
metics, faces need to be located and registered first to
facilitate further processing. It is evident that face detec-
tion plays an important and critical role for the success
of any face processing systems.
The face detection problem is challenging as it
needs to account for all possible appearance variation
caused by change in illu mination, facial features,
occlusions, etc. In addition, it has to detect faces that
appear at different scale, pose, with in- plane rotations.
In spite of all these difficulties, tremendous progress
has been made in the last decade and many systems
have shown impressive real-time performance. The
recent advances of these algorithms have also made
significant contributions in detecting other objects
such as humans/pedestrians, and cars.
Operation of a Face Detection System
Most detection systems carry out the task by extracting
certain properties (e.g., local features or holistic inten-
sity patterns) of a set of training images acquired at a
fixed pose (e.g., upright frontal pose) in an off-line
setting. To reduce the effects of illumination change,
these images are processed with histogram equalization
[1, 2] or standardization (i.e., zero mean unit variance)
[3]. Based on the extracted properties, these systems
typically scan through the entire image at every possi-
ble locatio n and scale in order to locate faces. The
extracted properties can be either manually coded
(with human knowledge) or learned from a set of
data as adopted in the recent systems that have demon-
strated impressive results [1, 2, 3, 4, 5]. In order to
detect faces at different scale, the detection process is
usually repeated to a pyramid of images whose resolu-
tion are reduced by a certain factor (e.g., 1.2) from the
original one [1, 2]. Such procedures may be exp edited
when other visual cues can be accurately incorporated
(e.g., color and motion) as pre-processing steps to
reduce the search space [5]. As faces are often detected
across scale, the raw detected faces are usually further
processed to combine overlapped results and remove
false positives with heuristics (e.g., faces typically do
not overlap in images) [2] or further processing (e.g.,
edge detection and intensity variance).
Numerous representations have been proposed for
face detection, including pixel-based [1, 2, 5], parts-
based [4, 6, 7], local edge features [8], Haar wavelets
[4, 9], and
▶ Haar-like features [3, 10 ]. While earlier
holistic representation schemes are able to detect faces
[1, 2, 5], the recent systems with Haar-like features
[3, 11, 12] have demonstrated impressive empirical
results in detecting faces under occlusion. A large and
representative training set of face images is essential for
the success of learning-based face detectors. From the
set of collected data, more positive examples can be
synthetically gen erated by perturbing, mirroring, ro-
tating, and scaling the original face images [1, 2].
On the other hand, it is relatively easier to collect
negative examples by randomly samplin g images with-
out face images [1, 2].
As face detection can be mainly formulated as a
pattern recognition problem, numerous algorithms
have been proposed to learn their generic templates
(e.g., eigenface and statistical distribution) or dis-
criminant class ifiers (e.g., neural networks, Fisher
linear discriminant, sparse network of Winnows, de-
cision tree, Bayes classifiers, support vector machines,
and
▶ AdaBoost). Typically, a good face detection
system needs to be trained with several iterations.
One common method to further improve the system
is to bootstrap a trained face detector with test sets,
and retrain the system with the false positive as well as
negatives [2]. This process is repeated several times to
further improve the performance of a face detector.
A survey on these topics can be found in [5], and
the most recent advances are discussed in the next
section.
Recent Advances
The AdaBoost-based face detector by Viola and Jones
[3] demonstrated that faces can be fairly reliably
detected in real-time (i.e., more than 15 frames per
second on 320 240 images with desktop computers)
under partial occlusion. While Haar wavelets were
used in [9] for representing faces and pedestrians,
they proposed the use of Haar-like features which
can be computed efficiently with integral image [3].
Figure 1 shows four types of Haar-like features that are
used to encode the horizontal, vertical, and diagonal
304
F
Face Detection
intensity information of face images at different posi-
tion and scale.
Given a sample image of 24 24 pixels, the ex-
haustive set of parameterized Haar-like features
(at different position and scale) is very lar ge (about
160,000). Contrary to most of the prior algorithms
that use one single strong classifier (e.g., neural net-
works and support vector machines), they used an
ensemble of weak classifiers where each one is con-
structed by thresholding of one Haar-like feature. The
weak classifier s are selected and weighted using the
AdaBoost algorithm [13]. It is worth to note that
boosting algorithms can also be derived from the per-
spective of function approximation with gradient de-
scent and applications for regression [14]. As there are
large number of weak classifiers, they presented a
method to rank these classifiers into several cascades
using a set of optimization criteria. Within each stage,
an ensemble of several weak classifiers is trained using
the AdaBoost algorithm. The motivation behind the
cascade of classifier is that simple classifiers at early
stage can filter out most negative example s efficiently,
and stronger classifiers at later stage are only necessary
to deal with instances that look like faces. The final
detector, a 38 layer cascade of classifiers with 6,060
Haar-like features, demonstrated impressive real-time
performance with fairly high detection and low false
positive rates. Several extensions to detect faces in
multiple views with in-plane ration have since been
proposed [11, 12, 15]. An implementation of the Ada-
Boost-based face detector [3] can be found in the Intel
OpenCV library.
Despite the excellent run-time performance of
boosted cascade classifier [3], the training time of
such a system is rather lengthy. In addition, the
▶ clas-
sifier cascade is an example of degenerate decision tree
with an unbalanced data set (i.e., a small set of positive
examples and a huge set of negative ones). Numerous
algorithms have been proposed to address these issues
and extended to detect faces in multiple views. To
handle the asymmetry between the positive and nega-
tive data sets, Viola and Jones proposed the asymmet-
ric AdaBoost algorithm [16] which keeps most of the
weights on the positive examples. In [3], the AdaBoost
algorithm is used to select a specifie d number of weak
classifiers with lowest error rates for each cascade and
the process is repeated until a set of optimization
criteria (i.e., the number of stages, the number of
features of each stage, and the detection/false positive
rates) is satisfied. As each weak classifier is mad e of one
single Haar-like feature, the process within each stage
can be considered as a feature selection problem. In-
stead of repeating the feature selection process at each
stage, Wu et al. [17] presented a greedy algorithm for
determining the set of features for all stages first before
training the cascade classifier. With the greedy feature
selection algorithm used as a pre-computing proce-
dure, they reported that the training time of the classi-
fier cascade with AdaBoost is reduced by 50–100 times.
For learning in each stage (or node) within the classifier
cascade, they also exploited the asymmetry between
positive and negative data using a linear classifier with
the assumption that they can be modeled with Gaussian
distributions [17]. The merits a nd dr awbacks of the
Face Detection. Figure 1 Four types of Haar-like features. These features appear at different position and scale. The
Haar-like features are computed as the difference of dark and light regions. They can be considered as features that
collect local edge information at different orientation and scale. The set of Haar-like features is large, and only a small
amount of them are learned from positive and negative examples for face detection.
Face Detection
F
305
F
proposed linear asymmetric classifier as well as the
classic Fisher linear discriminant were also examined
in their work. Recently, Pham and Cham proposed an
online algorithm that learns asymmetric boosted clas-
sifiers [18] with significant gain in training time.
In [19], an algorithm that aims to automatically
determine the number of classifiers and stages for
constructing a boosted ensemble was proposed.
While a greedy optimization algorithm was employed
in [3], Brubaker et al. proposed an algorithm for de-
termining the number of weak classifiers and training
each node classifier of a cascade by selecting operat-
ing points within a receiver operator characteristic
(ROC) curve [20]. The solved optimization problem
using linear programs that maximize the detection
rates while satisfying the constraints of false positive
rates [19].
Although the original four types of Haar-like fea-
tures are sufficient to encode upright frontal face
images, other types of features are essential to represent
more complex patterns (e.g., faces in different pose)
[10, 11, 12, 15]. Most systems take a divide-and-
conquer strategy and a face detector is constructed
for a fixed pose, thereby covering a wide range of angles
(e.g., yaw and pitch angles). A test image is either
sent to all detectors for evalu ation, or to a decision
module with a coarse pose estimator for selecting the
appropriate trees for further processing. The ensuing
problems are how the types of features are constructed,
and how the most important ones from a large feature
space are selected. More generalized Haar-like features
are defined in [10, 11] in which the rectangular image
regions are not necessarily adjacent, and furthermore
the number of such rectangular blocks is randomly
varied [10]. Several greedy algorithms have been pro-
posed to select features efficien tly by exploiting the
statistics of features before training boosted cascade
classifiers [17].
There are also other fast face detection methods that
demonstrate promising results, including the compo-
nent-based face detector using Naive Bayes classifiers
[4], the face detectors using support vector machines
[7, 21, 22], the Anti-face method [23] which consists of
a series of detectors trained with positive images only,
and the energy-based method [24] that simultaneously
detects faces and estimates their pose in real time.
Quantifying Performance
There are numerous metrics to gauge the performance
of face detection systems, ranging from detection
frame rate, false positive/negative rate, number of classi-
fier, number of feature, number of training image, train-
ing time, accuracy, and memory requirements. In
addition, the reported performance also depends on the
definition of a ‘‘correct’’ detection result [2, 5]. Figure 2
shows the effects of detection results versus different
criteria, and more discussions can be found in [2, 5].
The most commonly adopted method is to plot the
▶ ROC cur ve using the de facto standard MIT + CMU
data set [2] which contains frontal face images. Anoth-
er data set from CMU contains images with faces that
vary in pose from frontal to side view [4]. Note that
Face Detection. Figure 2 Detection results depend heavily on the adopted criteria. Suppose all the sub-images in (b) are
306
F
Face Detection
although the face detection methods nowadays have
impressive real-time performance, there is still much
room for improvement in terms of accuracy. The
detected faces returned by state-of-the-art algorithms
are often a few pixels (around 5) off the ‘‘accurate’’
locations, which is significant as face imag es are usually
standardized to 21 21 pixels. While such results are
the trade-offs between speed, robustness, and accuracy,
they inevitably degrade the performance of any bio-
metric applications using the contents of detected
faces. Several post-processing algorithms have been
proposed to better locate faces and extract facial fea-
tures (when the image resolution of the detected faces
is sufficiently high) [25].
Applications
As face detection is the first step of any face processing
system, it finds numerous applications in face rec ogni-
tion, face tracking, facial expression rec ognition, facial
feature extraction, gender classification, clustering, atten-
tive user interfaces, digital cosmetics, biometric systems,
to name a few. In addition, most of the face detection
algorithms can be extended to r ecognize other objects
such as cars, humans, pedestrian s, and signs, etc. [5].
Summary
Recent advances in face detection have created a lot of
exciting and reasonably robust applications. As most of
the developed algorithms can also be applied to other
problem domains, it has broader impact than detecting
faces in images alone. Future research will focus on
improvement of detection precision (in terms of loca-
tion), online training of such detectors, and novel
applications.
Related Entries
▶ Biometric Algorithm
▶ Ensemble Learning
▶ Face Tracking
▶ Face Recognition, Over view
▶ Facial Expression Recognition
▶ Machine-Learning
▶ Supervised Learning Surveillance
References
1. Sung, K.K., Poggio, T.: Example-based learning for view-based
human face detection. IEEE Trans. Pattern Anal. Mach. Intell.
20(1), 39–51 (1998)
2. Rowley, H., Baluja, S., Kanade, T.: Neural network-based face
detection. IEEE Trans. Pattern Anal. Mach. Intell. 20(1), 23–28
(1998)
3. Viola, P., Jones, M.: Robust real-time face detection. Int.
J. Comput. Vision 57(2), 137–154 (2004)
4. Schneiderman, H., Kanade, T.: Object detection using the statis-
tics of parts. Int. J. Comput. Vision 56(3), 151–177 (2004)
5. Yang, M.H., Kriegman, D., Ahuja, N.: Detecting faces in images:
A survey. IEEE Trans. Pattern Anal. Mach. Intell. 24(1), 34–58
(2002)
6. Mohan, A., Papageorgiou, C., Poggio, T.: Example-based object
detection in images by components. IEEE Trans. Pattern Anal.
Mach. Intell. 23(4), 349–361 (2001)
7. Heisele, B., Serre, T., Poggio, T.: A component-based framework
for face detection and identification. Int. J. Comput. Vision
74(2), 167–181 (2007)
8. Fleuret, F., Geman, D.: Coarse-to-fine face detection. Int.
J. Comput. Vision 41(12), 85–107 (2001)
9. Papageorgiou, C., Poggio, T.: A trainable system for object rec-
ognition. Int. J. Comput. Vision 38(1), 15–33 (2000)
10. Dollar, P., Tu, Z., Tao, H., Belongie, S.: Feature mining for image
classification. In: Proceedings of IEEE Conference on Computer
Vision and Pattern Recognition (2007)
11. Li, S., Zhang, Z.: Floatboost learning and statistical face detec-
tion. IEEE Trans. Pattern Anal. Mach. Intell. 28(9), 1112–1123
(2004)
12. Huang, C., Ai, H., Li, Y., Lao, S.: High-performance rotation
invariant multiview face detection. IEEE Trans. Pattern Anal.
Mach. Intell. 29(4), 671–686 (2007)
13. Freund, Y., Schapire, R.: A decision-theoretic generalization of
on-line learning and application to boosting. J. Comput. Syst.
Sci. 55(1), 119–139 (1997)
14. Friedman, J., Hastie, T., Tibshirani, R.: Additive logistic regres-
sion: a statistical view of boosting (With discussion and a rejoin-
der by the authors). Ann. Stat. 28(2), 337–407 (2000)
15. Jones, M., Viola, P.: Fast multi-view face detection. Technical
Report TR2003-96, Mitsubishi Electrical Research Laboratories
(2003)
16. Viola, P., Jones, M.: Fast and robust classification using asym-
metric Adaboost and a detector cascade. In: Advances in Neural
Information Processing Systems, pp. 1311–1318 (2002)
17. Wu, J., Brubaker, S.C., Mullin, M., Rehg, J.: Fast asymmetric
learning for cascade face detection. IEEE Trans. Pattern Anal.
Mach. Intell. 30(3), 369–382 (2008)
18. Pham, M.T., Cham, T.J.: Online learning asymmetric boosted
classifiers for object detection. In: Proceedings of IEEE Confer-
ence on Computer Vision and Pattern Recognition (2007)
19. Brubaker, S.C., Wu, J., Sun, J., Mullin, M., Rehg, J.: On the design
of cascades of boosted ensembles for face detection. Int.
J. Comput. Vision 77(1–3), 65–86 (2008)
20. Provost, F., Fawcett, T.: Robust classification for imprecise envir-
onments. Mach. Learn. 42(3), 203–231 (2001)
Face Detection
F
307
F
21. Oren, M., Papageorgiou, C., Sinha, P., Osuna, E., Poggio, T.:
Pedestrian detection using wavelet templates. In: Proceedings of
IEEE Conference on Computer Vision and Pattern Recognition,
pp. 193–199 (1997)
22. Romdhani, S., Torr, P., Scho
¨
lkopf, B., Blake, A.: Computationally
efficient face detection. In: Proceedings of the Eighth IEEE
International Conference on Computer Vision, vol. 2, pp. 695–
700 (2001)
23. Keren, D., Osadchy, M., Gotsman, C.: Antifaces: A novel fast
method for image detection. IEEE Trans. Pattern Anal. Mach.
Intell. 23(7), 747–761 (2001)
24. Osadchy, M., LeCun, Y., Miller, M.: Synergistic face detection
and pose estimation with energy-based models. J. Mach. Learn.
Res. 1197–1214 (2007)
25. Ding, L., Martinez, A.: Precise detailed detection of faces and
facial features. In: Proceedings of IEEE Conference on Computer
Vision and Pattern Recognition (2008)
Face Device
MASSIMO TISTARELLI
Computer Vision Laboratory, University of Sassari,
Piazza Duomo, 6 Alghero, Italy
Synonyms
Face acquisition; Face camera; Video camera; Visual
sensor
Definition
A face device is a system to acquire a set of digital data
samples representing a human face. As the human face
is a complex 3D object, the data can be in several
forms: a 2D image where the gray levels of the
▶ pixels
represent the projected reflectance of the face surface
under visible illumination; a 2D image where the gray
levels of the pixels represent the projected reflectance
of the face surface illuminated with an active source;
a 2D thermal image representing the heat emitted by
the face surface; 3D samples of the surface structure.
Face devices can be distinguished on the basis of
the data dimension if it is active or passive. Face
devices can be passive, i.e., based on the passive reflec-
tance of ambient light by the body, or active, i.e.,
associated with an energy emitte r and a sensor to
capture the energy reflected by the face. The data
captured can be either in 2D or 3D form.
A face device can be based on different technolo-
gies, depending upon the data to be captured and the
signal to be obtained. The most applied face devices
include a video
▶ camera to capture 2D images of the
face and a digitiz er to sample and quantize the analog
signal generated by the camera. Different face devices
deliver different signals to be digitized into 2D or 3D
data. The data captured can be stored under different
file formats for subsequent processing.
Introduction
Current face biometric systems are based on the acqui-
sition and processing of image data, representing a
human face. A face acquisition device is typically
a video camera capable of acquiring single images or
video streams of data, representing a face. As the face is
a 3D object, the acquired data can represent either the
2D projection of the face reflectance on the image
plane or a set of 3D samples of the face structure,
possibly with the associated reflectance. In the former
case, a conventional video camera can be used to
acquire images of face. In the latter case, a more com-
plex 3D acquisition device must be applied.
2D Face Devices
A conventional camera acquires the image data as a
reflectance of the imaged scene. The face points are
recorded as the geometrical projection of the 3D points
on the face surface onto the 2D image plane (Fig. 1).
Several video cameras exist that are capable of
capturing either single images or video streams from
the viewed scene. The most critical parts of the camera
are the acquisition sensor and the lenses.
Charge-coupled dev ice (CCD) and complementary
metal oxide semiconductor (CMOS) image sensors
are two different technologies for capturing images
digitally; current commercial camera adopt either of
these. Both types of sensors convert light into electric
charge and process it into electronic signals. In a
▶ CCD sensor, charge of every pixel is transferred
through a very limited num ber of outp ut nodes
308
F
Face Device
(often just one) to be converted to voltage, buffered,
and sent off-chip as an analog signal. All the pixel can
be devoted to capture light, and the output’s unifor-
mity (a key factor in image quality) is high. In a
▶ CMOS sensor, each pixel has its own charge-to-
voltage conversion, and the sensor often includes
amplifiers, noise-correction, and digitization circuits,
so that the chip outputs digital bits. These other func-
tions increase the design complexity and reduce the
area available for light capture. With each pixel doing
its own conversion, the uniformity is low. But the chip
can be built to require less off-chip circuitry for basic
operation.
The CMOS pixel solves the speed and scalability
issues of the CCD sensor. They consume far less power
than a CCD, have less image lag, and can be fabricated
on much cheaper and more available manufacturing
lines. Unlike CCDs, CMOS sensors can combine both
the image sensor and the image processing functions
within the same integrated circuit. CMOS imagers still
suffer from higher
▶ fixed-pattern noise than CCDs,
but active pixel sensors are catching up with respect to
noise, dynamic range, and responsivity. CMOS sensors
have become the technology of choice for many con-
sumer applications, most significantly, the burgeoning
cell phone camera market [1].
The technology of the sensor and the capturing
device determines several properties of the captured
signal such as the following:
1. Image resolution. This is related to both the active
elements on the imager sensor and the sampling
Face Device. Figure 1 Geometry of the pin hole camera model. (Top) 3D sketch of the projection of point P in space on
the image pixel Q. (Bottom) 2D projection of the Y–Z plane.
Face Device
F
309
F