Detecting Faces in Images: A Survey
Ming-Hsuan Yang, Member, IEEE, David J. Kriegman, Senior Member, IEEE,and
Narendra Ahuja, Fellow, IEEE
Abstract—Images containing faces are essential to intelligent vision-based human computer interaction, and research efforts in face
processing include face recognition, face tracking, pose estimation, and expression recognition. However, many reported methods
assume that the faces in an image or an image sequence have been identified and localized. To build fully automated systems that
analyze the information contained in face images, robust and efficient face detection algorithms are required. Given a single image, the
goal of face detection is to identify all image regions which contain a face regardless of its three-dimensional position, orientation, and
lighting conditions. Such a problem is challenging because faces are nonrigid and have a high degree of variability in size, shape, color,
and texture. Numerous techniques have been developed to detect faces in a single image, and the purpose of this paper is to
categorize and evaluate these algorithms. We also discuss relevant issues such as data collection, evaluation metrics, and
benchmarking. After analyzing these algorithms and identifying their limitations, we conclude with several promising directions for
Index Terms—Face detection, face recognition, object recognition, view-based recognition, statistical pattern recognition, machine
ITH the ubiquity of new information technology and
media, more effective and friendly methods for
human computer interaction (HCI) are being developed
which do not rely on traditional devices such as keyboards,
mice, and displays. Furthermore, the ever decreasing price/
performance ratio of computing coupled with recent
decreases in video image acquisition cost imply that
computer vision systems can be deployed in desktop and
embedded systems , , . The rapidly expand-
ing research in face processing is based on the premise that
information about a user’s identity, state, and intent can be
extracted from images, and that computers can then react
accordingly, e.g., by observing a person’s facial expression.
In the last five years, face and facial expression recognition
have attracted much attention though they have been
studied for more than 20 years by psychophysicists,
neuroscientists, and engineers. Many research demonstra-
tions and commercial applications have been developed
from these efforts. A first step of any face processing system
is detecting the locations in images where faces are present.
However, face detection from a single image is a challen-
ging task because of variability in scale, location, orientation
(up-right, rotated), and pose (frontal, profile). Facial
expression, occlusion, and lighting conditions also change
the overall appearance of faces.
We now give a definition of face detection: Given an
arbitrary image, the goal of face detection is to determine
whether or not there are any faces in the image and, if
present, return the image location and extent of each face.
The challenges associated with face detection can be
attributed to the following factors:
. Pose. The images of a face vary due to the relative
camera-face pose (frontal, 45 degree, profile, upside
down), and some facial features such as an eye or the
nose may become partially or wholly occluded.
. Presence or absence of structural components.
Facial features such as beards, mustaches, and
glasses may or may not be present and there is a
great deal of variability among these components
including shape, color, and size.
. Facial expression. The appearance of faces are
directly affected by a person’s facial expression.
. Occlusion. Faces may be partially occluded by other
objects. In an image with a group of people, some
faces may partially occlude other faces.
. Image orientation. Face images directly vary for
different rotations about the camera’s optical axis.
. Imaging conditions. When the image is formed,
factors such as lighting (spectra, source distribution
and intensity) and camera characteristics (sensor
response, lenses) affect the appearance of a face.
There are many closely related problems of face
detection. Face localization aims to determine the image
position of a single face; this is a simplified detection
problem with the assumption that an input image contains
only one face , . The goal of facial feature detection is
to detect the presence and location of features, such as eyes,
nose, nostrils, eyebrow, mouth, lips, ears, etc., with the
assumption that there is only one face in an image , .
Face recognition or face identification compares an input image
(probe) against a database (gallery) and reports a match, if
34 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
. M.-H. Yang is with Honda Fundamental Research Labs, 800 California
Street, Mountain View, CA 94041. E-mail: email@example.com.
. D.J. Kriegman is with the Department of Computer Science and Beckman
Institute, University of Illinois at Urbana-Champaign, Urbana, IL 61801.
. N. Ahjua is with the Department of Electrical and Computer Engineering
and Beckman Institute, University of Illinois at Urbana-Champaign,
Urbana, IL 61801. E-mail: firstname.lastname@example.org.
Manuscript received 5 May 2000; revised 15 Jan. 2001; accepted 7 Mar. 2001.
Recommended for acceptance by K. Bowyer.
For information on obtaining reprints of this article, please send e-mail to:
email@example.com, and reference IEEECS Log Number 112058.
0162-8828/02/$17.00 ß 2002 IEEE
any , , . The purpose of face authentication is to
verify the claim of the identity of an individual in an input
image , , while face tracking methods continuously
estimate the location and possibly the orientation of a face
in an image sequence in real time , , . Facial
expression recognition concerns identifying the affective
states (happy, sad, disgusted, etc.) of humans , .
Evidently, face detection is the first step in any automated
system which solves the above problems. It is worth
mentioning that many papers use the term “face detection,”
but the methods and the experimental results only show
that a single face is localized in an input image. In this
paper, we differentiate face detection from face localization
since the latter is a simplified problem of the former.
Meanwhile, we focus on face detection methods rather than
While numerous methods have been proposed to detect
faces in a single image of intensity or color images, we are
unaware of any surveys on this particular topic. A survey of
early face recognition methods before 1991 was written by
Samal and Iyengar . Chellapa et al. wrote a more recent
survey on face recognition and some detection methods .
Among the face detection methods, the ones based on
learning algorithms have attracted much attention recently
and have demonstrated excellent results. Since these data-
driven methods rely heavily on the training sets, we also
discuss several databases suitable for this task. A related
and important problem is how to evaluate the performance
of the proposed detection methods. Many recent face
detection papers compare the performance of several
methods, usually in terms of detection and false alarm
rates. It is also worth noticing that many metrics have been
adopted to evaluate algorithms, such as learning time,
execution time, the number of samples required in training,
and the ratio between detection rates and false alarms.
Evaluation becomes more difficult when researchers use
different definitions for detection and false alarm rates. In
this paper, detection rate is defined as the ratio between the
number of faces correctly detected and the number faces
determined by a human. An image region identified as a
face by a classifier is considered to be correctly detected if
the image region covers more than a certain percentage of a
face in the image (See Section 3.3 for details). In general,
detectors can make two types of errors: false negatives in
which faces are missed resulting in low detection rates and
false positives in which an image region is declared to be
face, but it is not. A fair evaluation should take these factors
into consideration since one can tune the parameters of
one’s method to increase the detection rates while also
increasing the number of false detections. In this paper, we
discuss the benchmarking data sets and the related issues in
a fair evaluation.
With over 150 reported approaches to face detection, the
research in face detection has broader implications for
computer vision research on object recognition. Nearly all
model-based or appearance-based approaches to 3D object
recognition have been limited to rigid objects while
attempting to robustly perform identification over a broad
range of camera locations and illumination conditions. Face
detection can be viewed as a two-class recognition problem
in which an image region is classified as being a “face” or
“nonface.” Consequently, face detection is one of the few
attempts to recognize from images (not abstract representa-
tions) a class of objects for which there is a great deal of
within-class variability (described previously). It is also one
of the few classes of objects for which this variability has
been captured using large training sets of images and, so,
some of the detection techniques may be applicable to a
much broader class of recognition problems.
Face detection also provides interesting challenges to the
underlying pattern classification and learning techniques.
When a raw or filtered image is considered as input to a
pattern classifier, the dimension of the feature space is
extremely large (i.e., the number of pixels in normalized
training images). The classes of face and nonface images are
decidedly characterized by multimodal distribution func-
tions and effective decision boundaries are likely to be
nonlinear in the image space. To be effective, either classifiers
must be able to extrapolate from a modest number of training
samples or be efficient when dealing with a very large
number of these high-dimensional training samples.
With an aim to give a comprehensive and critical survey
of current face detection methods, this paper is organized as
follows: In Section 2, we give a detailed review of
techniques to detect faces in a single image. Benchmarking
databases and evaluation criteria are discussed in Section 3.
We conclude this paper with a discussion of several
promising directions for face detection in Section 4.
Though we report error rates for each method when
available, tests are often done on unique data sets and, so,
comparisons are often difficult. We indicate those methods
that have been evaluated with a publicly available test set. It
can be assumed that a unique data set was used if we do not
indicate the name of the test set.
2DETECTING FACES IN A SINGLE IMAGE
In this section, we review existing techniques to detect faces
from a single intensity or color image. We classify single
image detection methods into four categories; some
methods clearly overlap category boundaries and are
discussed at the end of this section.
1. Knowledge-based methods. These rule-based meth-
ods encode human knowledge of what constitutes a
typical face. Usually, the rules capture the relation-
ships between facial features. These methods are
designed mainly for face localization.
2. Feature invariant approaches. These algorithms aim
to find structural features that exist even when the
pose, viewpoint, or lighting conditions vary, and
then use the these to locate faces. These methods are
designed mainly for face localization.
3. Template matching methods. Several standard pat-
terns of a face are stored to describe the face as a whole
or the facial features separately. The correlations
between an input image and the stored patterns are
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 35
1. An earlier version of this survey paper appeared at http://
vision.ai.uiuc.edu/mhyang/face-dectection-survey.html in March 1999.
computed for detection. These methods have been
used for both face localization and detection.
4. Appearance-based methods. In contrast to template
matching, the models (or templates) are learned from
a set of training images which should capture the
representative variability of facial appearance. These
learned models are then used for detection. These
methods are designed mainly for face detection.
Table 1 summarizes algorithms and representative
works for face detection in a single image within these
four categories. Below, we discuss the motivation and
general approach of each category. This is followed by a
review of specific methods including a discussion of their
pros and cons. We suggest ways to further improve these
methods in Section 4.
2.1 Knowledge-Based Top-Down Methods
In this approach, face detection methods are developed
based on the rules derived from the researcher’s knowledge
of human faces. It is easy to come up with simple rules to
describe the features of a face and their relationships. For
example, a face often appears in an image with two eyes
that are symmetric to each other, a nose, and a mouth. The
relationships between features can be represented by their
relative distances and positions. Facial features in an input
image are extracted first, and face candidates are identified
based on the coded rules. A verification process is usually
applied to reduce false detections.
One problem with this approach is the difficulty in
translating human knowledge into well-defined rules. If the
rules are detailed (i.e., strict), they may fail to detect faces
that do not pass all the rules. If the rules are too general,
they may give many false positives. Moreover, it is difficult
to extend this approach to detect faces in different poses
since it is challenging to enumerate all possible cases. On
the other hand, heuristics about faces work well in detecting
frontal faces in uncluttered scenes.
Yang and Huang used a hierarchical knowledge-based
method to detect faces . Their system consists of three
levels of rules. At the highest level, all possible face
candidates are found by scanning a window over the input
image and applying a set of rules at each location. The rules
at a higher level are general descriptions of what a face
looks like while the rules at lower levels rely on details of
facial features. A multiresolution hierarchy of images is
created by averaging and subsampling, and an example is
shown in Fig. 1. Examples of the coded rules used to locate
face candidates in the lowest resolution include: “the center
36 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
Categorization of Methods for Face Detection in a Single Image
Fig. 1. (a) n = 1, original image. (b) n = 4. (c) n = 8. (d) n = 16. Original and corresponding low resolution images. Each square cell consists of
n n pixels in which the intensity of each pixel is replaced by the average intensity of the pixels in that cell.
part of the face (the dark shaded parts in Fig. 2) has four
cells with a basically uniform intensity,” “the upper round
part of a face (the light shaded parts in Fig. 2) has a basically
uniform intensity,” and “the difference between the average
gray values of the center part and the upper round part is
significant.” The lowest resolution (Level 1) image is
searched for face candidates and these are further processed
at finer resolutions. At Level 2, local histogram equalization
is performed on the face candidates received from Level 2,
followed by edge detection. Surviving candidate regions are
then examined at Level 3 with another set of rules that
respond to facial features such as the eyes and mouth.
Evaluated on a test set of 60 images, this system located
faces in 50 of the test images while there are 28 images in
which false alarms appear. One attractive feature of this
method is that a coarse-to-fine or focus-of-attention strategy
is used to reduce the required computation. Although it
does not result in a high detection rate, the ideas of using a
multiresolution hierarchy and rules to guide searches have
been used in later face detection works .
Kotropoulos and Pitas  presented a rule-based
localization method which is similar to  and . First,
facial features are located with a projection method that
Kanade successfully used tolocate the boundary of a face .
Let Iðx; yÞ be the intensity value of an m n image at position
ðx; yÞ, the horizontal and vertical projections of the image are
defined as HIðxÞ¼
Iðx; yÞ and VIðyÞ¼
The horizontal profile of an input image is obtained first, and
then the two local minima, determined by detecting abrupt
changes in HI, are said to correspond to the left and right side
of the head. Similarly, the vertical profile is obtained and the
local minima are determined for the locations of mouth lips,
nose tip, and eyes. These detected features constitute a facial
candidate. Fig. 3a shows one example where the boundaries
of the face correspond to the local minimum where abrupt
intensity changes occur. Subsequently, eyebrow/eyes, nos-
trils/nose, and the mouth detection rules are used to validate
these candidates. The proposed method has been tested using
a set of faces in frontal views extracted from the European
ACTS M2VTS (MultiModal Verification for Teleservices and
Security applications) database  which contains video
sequences of 37 different people. Each image sequence
contains only one face in a uniform background. Their
method provides correct face candidates in all tests. The
detection rate is 86.5 percent if successful detection is defined
as correctly identifying all facial features. Fig. 3b shows one
example in which it becomes difficult to locate a face in a
complex background using the horizontal and vertical
profiles. Furthermore, this method cannot readily detect
multiple faces as illustrated in Fig. 3c. Essentially, the
projection method can be effective if the window over
which it operates is suitably located to avoid misleading
2.2 Bottom-Up Feature-Based Methods
In contrast to the knowledge-based top-down approach,
researchers have been trying to find invariant features of
faces for detection. The underlying assumption is based on
the observation that humans can effortlessly detect faces
and objects in different poses and lighting conditions and,
so, there must exist properties or features which are
invariant over these variabilities. Numerous methods have
been proposed to first detect facial features and then to infer
the presence of a face. Facial features such as eyebrows,
eyes, nose, mouth, and hair-line are commonly extracted
using edge detectors. Based on the extracted features, a
statistical model is built to describe their relationships and
to verify the existence of a face. One problem with these
feature-based algorithms is that the image features can be
severely corrupted due to illumination, noise, and occlu-
sion. Feature boundaries can be weakened for faces, while
shadows can cause numerous strong edges which together
render perceptual grouping algorithms useless.
2.2.1 Facial Features
Sirohey proposed a localization method to segment a face
from a cluttered background for face identification . It
uses an edge map (Canny detector ) and heuristics to
remove and group edges so that only the ones on the face
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 37
Fig. 2. A typical face used in knowledge-based top-down methods:
Rules are coded based on human knowledge about the characteristics
(e.g., intensity distribution and difference) of the facial regions .
Fig. 3. (a) and (b) n = 8. (c) n = 4. Horizontal and vertical profiles. It is feasible to detect a single face by searching for the peaks in horizontal and
vertical profiles. However, the same method has difficulty detecting faces in complex backgrounds or multiple faces as shown in (b) and (c).
contour are preserved. An ellipse is then fit to the boundary
between the head region and the background. This algorithm
achieves 80 percent accuracy on a database of 48 images with
cluttered backgrounds. Instead of using edges, Chetverikov
and Lerch presented a simple face detection method using
blobs and streaks (linear sequences of similarly oriented
edges) . Their face model consists of two dark blobs and
three light blobs to represent eyes, cheekbones, and nose. The
model uses streaks to represent the outlines of the faces,
eyebrows, and lips. Two triangular configurations are
utilized to encode the spatial relationship among the blobs.
A low resolution Laplacian image is generated to facilitate
blob detection. Next, the image is scanned to find specific
triangular occurrences as candidates. A face is detected if
streaks are identified around a candidate.
Graf et al. developed a method to locate facial features
and faces in gray scale images . After band pass
filtering, morphological operations are applied to enhance
regions with high intensity that have certain shapes (e.g.,
eyes). The histogram of the processed image typically
exhibits a prominent peak. Based on the peak value and its
width, adaptive threshold values are selected in order to
generate two binarized images. Connected components are
identified in both binarized images to identify the areas of
candidate facial features. Combinations of such areas are
then evaluated with classifiers, to determine whether and
where a face is present. Their method has been tested with
head-shoulder images of 40 individuals and with five video
sequences where each sequence consists of 100 to
200 frames. However, it is not clear how morphological
operations are performed and how the candidate facial
features are combined to locate a face.
Leung et al. developed a probabilistic method to locate a
face in a cluttered scene based on local feature detectors and
random graph matching . Their motivation is to formulate
the face localization problem as a search problem in which the
goal is to find the arrangement of certain facial features that is
most likely to be a face pattern. Five features (two eyes, two
nostrils, and nose/lip junction) are used to describe a typical
face. For any pair of facial features of the same type (e.g., left-
eye, right-eye pair), their relative distance is computed, and
over an ensemble of images the distances are modeled by a
Gaussian distribution. A facial template is defined by
averaging the responses to a set of multiorientation, multi-
scale Gaussian derivative filters (at the pixels inside the facial
feature) over a number of faces in a data set. Given a test
image, candidate facial features are identified by matching
the filter response at each pixel against a template vector of
responses (similar to correlation in spirit). The top two feature
candidates with the strongest response are selected to search
for the other facial features. Since the facial features cannot
appear in arbitrary arrangements, the expected locations of
the other features are estimated using a statistical model of
mutual distances. Furthermore, the covariance of the esti-
mates can be computed. Thus, the expected feature locations
can be estimated with high probability. Constellations are
then formed only from candidates that lie inside the
appropriate locations, and the most face-like constellation is
determined. Finding the best constellation is formulated as a
random graph matching problem in which the nodes of the
graph correspond to features on a face, and the arcs represent
the distances between different features. Ranking of
constellations is based on a probability density function that
a constellation corresponds to a face versus the probability it
was generated by an alternative mechanism (i.e., nonface).
They used a set of 150 images for experiments in which a face
is considered correctly detected if any constellation correctly
locates three or more features on the faces. This system is able
to achieve a correct localization rate of 86 percent.
Instead of using mutual distances to describe the
relationships between facial features in constellations, an
alternative method for modeling faces was also proposed
by the Leung et al. , . The representation and
ranking of the constellations is accomplished using the
statistical theory of shape, developed by Kendall  and
Mardia and Dryden . The shape statistics is a joint
probability density function over N feature points, repre-
sented by ðx
Þ, for the ith feature under the assumption
that the original feature points are positioned in the plane
according to a general 2N-dimensional Gaussian distribu-
tion. They applied the same maximum-likelihood (ML)
method to determine the location of a face. One advantage
of these methods is that partially occluded faces can be
located. However, it is unclear whether these methods can
be adapted to detect multiple faces effectively in a scene.
In , , Yow and Cipolla presented a feature-
based method that uses a large amount of evidence from the
visual image and their contextual evidence. The first stage
applies a second derivative Gaussian filter, elongated at an
aspect ratio of three to one, to a raw image. Interest points,
detected at the local maxima in the filter response, indicate
the possible locations of facial features. The second stage
examines the edges around these interest points and groups
them into regions. The perceptual grouping of edges is
based on their proximity and similarity in orientation and
strength. Measurements of a region’s characteristics, such as
edge length, edge strength, and intensity variance, are
computed and stored in a feature vector. From the training
data of facial features, the mean and covariance matrix of
each facial feature vector are computed. An image region
becomes a valid facial feature candidate if the Mahalanobis
distance between the corresponding feature vectors is
below a threshold. The labeled features are further grouped
based on model knowledge of where they should occur
with respect to each other. Each facial feature and grouping
is then evaluated using a Bayesian network. One attractive
aspect is that this method can detect faces at different
orientations and poses. The overall detection rate on a test
set of 110 images of faces with different scales, orientations,
and viewpoints is 85 percent . However, the reported
false detection rate is 28 percent and the implementation is
only effective for faces larger than 60 60 pixels. Subse-
quently, this approach has been enhanced with active
contour models , . Fig. 4 summarizes their feature-
based face detection method.
Takacs and Wechsler described a biologically motivated
face localization method based on a model of retinal feature
extraction and small oscillatory eye movements . Their
algorithm operates on the conspicuity map or region of
interest, with a retina lattice modeled after the magnocellular
38 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
ganglion cells in the human vision system. The first phase
computes a coarse scan of the image to estimate the location of
the face, based on the filter responses of receptive fields. Each
receptive field consists of a number of neurons which are
implemented with Gaussian filters tuned to specific orienta-
tions. The second phase refines the conspicuity map by
scanning the image area at a finer resolution to localize the
face. The error rate on a test set of 426 images (200 subjects
from the FERET database) is 4.69 percent.
Han et al. developed a morphology-based technique to
extract what they call eye-analogue segments for face
detection . They argue that eyes and eyebrows are the
most salient and stable features of human face and, thus,
useful for detection. They define eye-analogue segments as
edges on the contours of eyes. First, morphological
operations such as closing, clipped difference, and thresh-
olding are applied to extract pixels at which the intensity
values change significantly. These pixels become the eye-
analogue pixels in their approach. Then, a labeling process
is performed to generate the eye-analogue segments. These
segments are used to guide the search for potential face
regions with a geometrical combination of eyes, nose,
eyebrows and mouth. The candidate face regions are
further verified by a neural network similar to . Their
experiments demonstrate a 94 percent accuracy rate using a
test set of 122 images with 130 faces.
Recently, Amit et al. presented a method for shape
detection and applied it to detect frontal-view faces in still
intensity images . Detection follows two stages: focusing
and intensive classification. Focusing is based on spatial
arrangements of edge fragments extracted from a simple
edge detector using intensity difference. A rich family of such
spatial arrangements, invariant over a range of photometric
and geometric transformations, is defined. From a set of
300 training face images, particular spatial arrangements of
edges which are more common in faces than backgrounds are
selected using an inductive method developed in . Mean-
while, the CART algorithm  is applied to grow a
classification tree from the training images and a collection
of false positives identified from generic background images.
Given a test image, regions of interest are identified from the
spatial arrangements of edge fragments. Each region of
interest is then classified as face or background using the
learned CART tree. Their experimental results on a set of
100 images from the Olivetti (now AT&T) data set  report
a false positive rate of 0.2 percent per 1,000 pixels and a false
negative rate of 10 percent.
Human faces have a distinct texture that can be used to
separate them from different objects. Augusteijn and Skufca
developed a method that infers the presence of a face
through the identification of face-like textures . The
texture are computed using second-order statistical features
(SGLD)  on subimages of 16 16 pixels. Three types of
features are considered: skin, hair, and others. They used a
cascade correlation neural network  for supervised
classification of textures and a Kohonen self-organizing
feature map  to form clusters for different texture
classes. To infer the presence of a face from the texture
labels, they suggest using votes of the occurrence of hair
and skin textures. However, only the result of texture
classification is reported, not face localization or detection.
Dai and Nakano also applied SGLD model to face
detection . Color information is also incorporated with
the face-texture model. Using the face texture model, they
design a scanning scheme for face detection in color scenes
in which the orange-like parts including the face areas are
enhanced. One advantage of this approach is that it can
detect faces which are not upright or have features such as
beards and glasses. The reported detection rate is perfect for
a test set of 30 images with 60 faces.
2.2.3 Skin Color
Human skin color has been used and proven to be an
effective feature in many applications from face detection to
hand tracking. Although different people have different
skin color, several studies have shown that the major
difference lies largely between their intensity rather than
their chrominance , , . Several color spaces have
been utilized to label pixels as skin including RGB , ,
, normalized RGB , , , , , ,
, , , , HSV (or HSI) , , , ,
YCrCb , , YIQ , , YES , CIE XYZ ,
and CIE LUV .
Many methods have been proposed to build a skin color
model. The simplest model is to define a region of skin tone
pixels using Cr;Cb values , i.e., RðCr;CbÞ, from samples
of skin color pixels. With carefully chosen thresholds,
, a pixel is classified to have skin
tone if its values ðCr;CbÞ fall within the ranges, i.e., Cr
. Crowley and Coutaz used a
histogram hðr; gÞ of ðr; gÞ values in normalized RGB color
space to obtain the probability of obtaining a particular RGB-
vector given that the pixel observes skin , . In other
words, a pixel is classified to belong to skin color if hðr; gÞ,
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 39
Fig. 4. (a) Yow and Cipolla model a face as a plane with six oriented facial features (eyebrows, eyes, nose, and mouth) . (b) Each facial feature
is modeled as pairs of oriented edges. (c) The feature selection process starts with interest points, followed by edge detection and linking, and tested
by a statistical model (Courtesy of K.C. Yow and R. Cipolla).
where is a threshold selected empirically from the
histogram of samples. Saxe and Foulds proposed an iterative
skin identification method that uses histogram intersection in
HSV color space . An initial patch of skin color pixels,
called the control seed, is chosen by the user and is used to
initiate the iterative algorithm. To detect skin color regions,
their method moves through the image, one patch at a time,
and presents the control histogram and the current histogram
from the image for comparison. Histogram intersection 
is used to compare the control histogram and current
histogram. If the match score or number of instances in
common (i.e., intersection) is greater than a threshold, the
current patch is classified as being skin color. Kjeldsen and
Kender defined a color predicate in HSV color space to
separate skin regions from background  . In contrast to the
nonparametric methods mentioned above, Gaussian density
functions , ,  and a mixture of Gaussians , ,
 are often used to model skin color. The parameters in a
unimodal Gaussian distribution are often estimated using
maximum-likelihood , , . The motivation for
using a mixture of Gaussians is based on the observation that
the color histogram for the skin of people with different ethnic
background does not form a unimodal distribution, but
rather a multimodal distribution. The parameters in a
mixture of Gaussians are usually estimated using an
EM algorithm , . Recently, Jones and Rehg conducted
a large-scale experiment in which nearly 1 billion labeled skin
tone pixels are collected (in normalized RGB color space) .
Comparing the performance of histogram and mixture
models for skin detection, they find histogram models to be
superior in accuracy and computational cost.
Color information is an efficient tool for identifying facial
areas and specific facial features if the skin color model can be
properly adapted for different lighting environments. How-
ever, such skin color models are not effective where the
spectrum of the light source varies significantly. In other
words, color appearance is often unstable due to changes in
both background and foreground lighting. Though the color
constancy problem has been addressed through the formula-
tion of physics-based models , several approaches have
been proposed to use skin color in varying lighting
conditions. McKenna et al. presented an adaptive color
mixture model to track faces under varying illumination
conditions . Instead of relying on a skin color model based
on color constancy, they used a stochastic model to estimate
an object’s color distribution online and adapt to accom-
modate changes in the viewing and lighting conditions.
Preliminary results show that their system can track faces
within a range of illumination conditions. However, this
method cannot be applied to detect faces in a single image.
Skin color alone is usually not sufficient to detect or track
faces. Recently, several modular systems using a combina-
tion of shape analysis, color segmentation, and motion
information for locating or tracking heads and faces in an
image sequence have been developed , , , ,
. We review these methods in the next section.
2.2.4 Multiple Features
Recently, numerous methods that combine several facial
features have been proposed to locate or detect faces. Most of
them utilize global features such as skin color, size, and shape
to find face candidates, and then verify these candidates
using local, detailed features such as eye brows, nose, and
hair. A typical approach begins with the detection of skin-like
regions as described in Section 2.2.3. Next, skin-like pixels are
grouped together using connected component analysis or
clustering algorithms. If the shape of a connected region has
an elliptic or oval shape, it becomes a face candidate. Finally,
local features are used for verification. However, others, such
as , , have used different sets of features.
Yachida et al. presented a method to detect faces in color
images using fuzzy theory , , . They used two
fuzzy models to describe the distribution of skin and hair
color in CIE XYZ color space. Five (one frontal and four side
views) head-shape models are used to abstract the appear-
ance of faces in images. Each shape model is a 2D pattern
consisting of m n square cells where each cell may contain
several pixels. Two properties are assigned to each cell: the
skin proportion and the hair proportion, which indicate the
ratios of the skin area (or the hair area) within the cell to the
area of the cell. In a test image, each pixel is classified as hair,
face, hair/face, and hair/background based on the distribu-
tion models, thereby generating skin-like and hair-like
regions. The head shape models are then compared with the
extracted skin-like and hair-like regions in a test image. If they
are similar, the detected region becomes a face candidate. For
verification, eye-eyebrow and nose-mouth features are
extracted from a face candidate using horizontal edges.
Sobottka and Pitas proposed a method for face localization
and facial feature extraction using shape and color .
First, color segmentation in HSV space is performed to locate
skin-like regions. Connected components are then deter-
mined by region growing at a coarse resolution. For each
connected component, the best fit ellipse is computed using
geometric moments. Connected components that are well
approximated by an ellipse are selected as face candidates.
Subsequently, these candidates are verified by searching for
facial features inside of the connected components. Features,
such as eyes and mouths, are extracted based on the
observation that they are darker than the rest of a face. In
, , a Gaussian skin color model is used to classify
skin color pixels. To characterize the shape of the clusters in
the binary image, a set of 11 lowest-order geometric moments
is computed using Fourier and radial Mellin transforms. For
detection, a neural network is trained with the extracted
geometric moments. Their experiments show a detection rate
of 85 percent based on a test set of 100 images.
The symmetry of face patterns has also been applied to
face localization . Skin/nonskin classification is carried
out using the class-conditional density function in YES color
space followed by smoothing in order to yield contiguous
regions. Next, an elliptical face template is used to
determine the similarity of the skin color regions based on
Hausdorff distance . Finally, the eye centers are
localized using several cost functions which are designed
to take advantage of the inherent symmetries associated
with face and eye locations. The tip of the nose and the
center of the mouth are then located by utilizing the
distance between the eye centers. One drawback is that it is
effective only for a single frontal-view face and when both
40 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
eyes are visible. A similar method using color and local
symmetry was presented in .
In contrast to pixel-based methods, a detection method
based on structure, color, and geometry was proposed in
. First, multiscale segmentation  is performed to
extract homogeneous regions in an image. Using a Gaussian
skin color model, regions of skin tone are extracted and
grouped into ellipses. A face is detected if facial features
such as eyes and mouth exist within these elliptic regions.
Experimental results show that this method is able to detect
faces at different orientations with facial features such as
beard and glasses.
Kauth et al. proposed a blob representation to extract a
compact, structurally meaningful description of multispec-
tral satellite imagery . A feature vector at each pixel is
formed by concatenating the pixel’s image coordinates to
the pixel’s spectral (or textural) components; pixels are then
clustered using this feature vector to form coherent
connected regions, or “blobs.” To detect faces, each feature
vector consists of the image coordinates and normalized
chrominance, i.e., X ¼ðx; y;
Þ , . A
connectivity algorithm is then used to grow blobs and the
resulting skin blob whose size and shape is closest to that of
a canonical face is considered as a face.
Range and color have also been employed for face
detection by Kim et al. . Disparity maps are computed
and objects are segmented from the background with a
disparity histogram using the assumption that background
pixels have the same depth and they outnumber the pixels in
the foreground objects. Using a Gaussian distribution in
normalized RGB color space, segmented regions with a skin-
like color are classified as faces. A similar approach has been
proposed by Darrell et al. for face detection and tracking .
2.3 Template Matching
In template matching, a standard face pattern (usually
frontal) is manually predefined or parameterized by a
function. Given an input image, the correlation values with
the standard patterns are computed for the face contour,
eyes, nose, and mouth independently. The existence of a
face is determined based on the correlation values. This
approach has the advantage of being simple to implement.
However, it has proven to be inadequate for face detection
since it cannot effectively deal with variation in scale, pose,
and shape. Multiresolution, multiscale, subtemplates, and
deformable templates have subsequently been proposed to
achieve scale and shape invariance.
2.3.1 Predefined Templates
An early attempt to detect frontal faces in photographs is
reported by Sakai et al. . They used several subtemplates
for the eyes, nose, mouth, and face contour to model a face.
Each subtemplate is defined in terms of line segments. Lines
in the input image are extracted based on greatest gradient
change and then matched against the subtemplates. The
correlations between subimages and contour templates are
computed first to detect candidate locations of faces. Then,
matching with the other subtemplates is performed at the
candidate positions. In other words, the first phase deter-
mines focus of attention or region of interest and the second
phase examines the details to determine the existence of a
face. The idea of focus of attention and subtemplates has been
adopted by later works on face detection.
Craw et al. presented a localization method based on a
shape template of a frontal-view face (i.e., the outline shape
of a face) . A Sobel filter is first used to extract edges.
These edges are grouped together to search for the template
of a face based on several constraints. After the head
contour has been located, the same process is repeated at
different scales to locate features such as eyes, eyebrows,
and lips. Later, Craw et al. describe a localization method
using a set of 40 templates to search for facial features and a
control strategy to guide and assess the results from the
template-based feature detectors .
Govindaraju et al. presented a two stage face detection
method in which face hypotheses are generated and tested
, , . A face model is built in terms of features
defined by the edges. These features describe the curves of the
left side, the hair-line, and the right side of a frontal face. The
Marr-Hildreth edge operator is used to obtain an edge map of
an input image. A filter is then used to remove objects whose
contours are unlikely to be part of a face. Pairs of fragmented
contours are linked based on their proximity and relative
orientation. Corners are detected to segment the contour into
feature curves. These feature curves are then labeled by
checking their geometric properties and relative positions in
the neighborhood. Pairs of feature curves are joined by edges
if their attributes are compatible (i.e., if they could arise from
the same face). The ratios of the feature pairs forming an edge
is compared with the golden ratio and a cost is assigned to the
edge. If the cost of a group of three feature curves (with
different labels) is low, the group becomes a hypothesis.
When detecting faces in newspaper articles, collateral
information, which indicates the number of persons in the
image, is obtained from the caption of the input image to
select the best hypotheses . Their system reports a
detection rate of approximately 70 percent based on a test
set of 50 photographs. However, the faces must be upright,
unoccluded, and frontal. The same approach has been
extended by extracting edges in the wavelet domain by
Venkatraman and Govindaraju .
Tsukamoto et al. presented a qualitative model for face
pattern (QMF) , . In QMF, each sample image is
divided into a number of blocks, and qualitative features are
estimated for each block. To parameterize a face pattern,
“lightness” and “edgeness” are defined as the features in this
model. Consequently, this blocked template is used to
calculate “faceness” at every position of an input image. A
face is detected if the faceness measure is above a predefined
Silhouettes have also been used as templates for face
localization . A set of basis face silhouettes is obtained
using principal component analysis (PCA) on face examples
in which the silhouette is represented by an array of bits.
These eigen-silhouettes are then used with a generalized
Hough transform for localization. A localization method
based on multiple templates for facial components was
proposed in . Their method defines numerous hypoth-
eses for the possible appearances of facial features. A set of
hypotheses for the existence of a face is then defined in terms
of the hypotheses for facial components using the Dempster-
Shafer theory . Given an image, feature detectors compute
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 41
confidence factors for the existence of facial features. The
confidence factors are combined to determine the measures of
belief and disbelief about the existence of a face. Their system
is able to locate faces in 88 images out of 94 images.
Sinha used a small set of spatial image invariants to
describe the space of face patterns , . His key
insight for designing the invariant is that, while variations
in illumination change the individual brightness of different
parts of faces (such as eyes, cheeks, and forehead), the
relative brightness of these parts remain largely unchanged.
Determining pairwise ratios of the brightness of a few such
regions and retaining just the “directions” of these ratios
(i.e., Is one region brighter or darker than the other?)
provides a robust invariant. Thus, observed brightness
regularities are encoded as a ratio template which is a
coarse spatial template of a face with a few appropriately
chosen subregions that roughly correspond to key facial
features such as the eyes, cheeks, and forehead. The
brightness constraints between facial parts are captured
by an appropriate set of pairwise brighter-darker relation-
ships between subregions. A face is located if an image
satisfies all the pairwise brighter-darker constraints. The
idea of using intensity differences between local adjacent
regions has later been extended to a wavelet-based
representation for pedestrian, car, and face detection .
Sinha’s method has been extended and applied to face
localization in an active robot vision system , . Fig. 5
shows the enhanced template with 23 defined relations.
These defined relations are furthered classified into 11
essential relations (solid arrows) and 12 confirming rela-
tions (dashed arrows). Each arrow in the figure indicates a
relation, with the head of the arrow denoting the second
region (i.e., the denominator of the ratio). A relation is
satisfied for face temple if the ratio between two regions
exceeds a threshold and a face is localized if the number of
essential and confirming relations exceeds a threshold.
A hierarchical template matching method for face detec-
tion was proposed by Miao et al. . At the first stage, an
input image is rotated from 20
in steps of 5
, in order
to handle rotated faces. A multiresolution image hierarchy is
formed (See Fig. 1) and edges are extracted using the
Laplacian operator. The face template consists of the edges
produced by six facial components: two eyebrows, two eyes,
one nose, and one mouth. Finally, heuristics are applied to
determine the existence of a face. Their experimental results
show better results in images containing a single face (frontal
or rotated) than in images with multiple faces.
2.3.2 Deformable Templates
Yuille et al. used deformable templates to model facial
features that fit an a priori elastic model to facial features
(e.g.,eyes) . In this approach,facial featuresare described
by parameterized templates. An energy function is defined to
link edges, peaks, and valleys in the input image to
corresponding parameters in the template. The best fit of the
elasticmodel is found by minimizing anenergyfunction of the
parameters.Although their experimentalresults demonstrate
good performance in tracking nonrigid features, one draw-
back of this approach is that the deformable template must be
initialized in the proximity of the object of interest.
In , a detection method based on snakes ,  and
templates was developed. An image is first convolved with
a blurring filter and then a morphological operator to
enhance edges. A modified n-pixel (n is small) snake is used
to find and eliminate small curve segments. Each face is
approximated by an ellipse and a Hough transform of the
remaining snakelets is used to find a dominant ellipse.
Thus, sets of four parameters describing the ellipses are
obtained and used as candidates for face locations. For each
of these candidates, a method similar to the deformable
template method  is used to find detailed features. If a
substantial number of the facial features are found and if
their proportions satisfy ratio tests based on a face template,
a face is considered to be detected. Lam and Yan also used
snakes to locate the head boundaries with a greedy
algorithm in minimizing the energy function .
Lanitis et al. described a face representation method with
both shape and intensity information . They start with sets
of training images in which sampled contours such as the eye
boundary, nose, chin/cheek are manually labeled, and a
vector of sample points is used to represent shape. They used
a point distribution model (PDM) to characterize the shape
vectors over an ensemble of individuals, and an approach
similar to Kirby and Sirovich  to represent shape-
normalized intensity appearance. A face-shape PDM can be
used to locate faces in new images by using active shape
model (ASM) search to estimate the face location and shape
parameters. The face patch is then deformed to the average
shape, and intensity parameters are extracted. The shape and
intensity parameters can be used together for classification.
Cootes and Taylor applied a similar approach to localize a
face in an image . First, they define rectangular regions of
the image containing instances of the feature of interest.
Factor analysis  is then applied to fit these training features
and obtain a distribution function. Candidate features are
determined if the probabilistic measures are above a thresh-
old and are verified using the ASM. After training this
method with 40 images, it is able to locate 35 faces in 40 test
images. The ASM approach has also been extended with two
Kalman filters to estimate the shape-free intensity parameters
and to track faces in image sequences .
2.4 Appearance-Based Methods
Contrasted to the template matching methods where tem-
plates are predefined by experts, the “templates” in appear-
ance-based methods are learned from examples in images. In
general, appearance-based methods rely on techniques from
statistical analysis and machine learning to find the relevant
42 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
Fig. 5. A 14x16 pixel ratio template for face localization based on Sinha
method. The template is composed of 16 regions (the gray boxes) and
23 relations (shown by arrows)  (Courtesy of B. Scassellati).
characteristics of face and nonface images. The learned
characteristics are in the form of distribution models or
discriminant functions that are consequently used for face
detection. Meanwhile, dimensionality reduction is usually
carried out for the sake of computation efficiency and
Many appearance-based methods can be understood in a
probabilistic framework. An image or feature vector
derived from an image is viewed as a random variable x,
and this random variable is characterized for faces and
nonfaces by the class-conditional density functions
pðxjfaceÞ and pðxjnonfaceÞ. Bayesian classification or
maximum likelihood can be used to classify a candidate
image location as face or nonface. Unfortunately, a
straightforward implementation of Bayesian classification
is infeasible because of the high dimensionality of x ,
because pðxjfaceÞ and pðxjnonfaceÞ are multimodal, and
because it is not yet understood if there are natural
parameterized forms for pðxjfaceÞ and pðxjnonfaceÞ.
Hence, much of the work in an appearance-based method
concerns empirically validated parametric and nonpara-
metric approximations to pðxjfaceÞ and pðxjnonfaceÞ.
Another approach in appearance-based methods is to find
a discriminant function (i.e., decision surface, separating
hyperplane, threshold function) between face and nonface
classes. Conventionally, image patterns are projected to a
lower dimensional space and then a discriminant function is
formed (usually based on distance metrics) for classification
, or a nonlinear decision surface can be formed using
multilayer neural networks . Recently, support vector
machines and other kernel methods have been proposed.
These methods implicitly project patterns to a higher
dimensional space and then form a decision surface between
the projected face and nonface patterns .
An early example of employing eigenvectors in face
recognition was done by Kohonen  in which a simple
neural network is demonstrated to perform face recognition
for aligned and normalized face images. The neural
network computes a face description by approximating
the eigenvectors of the image’s autocorrelation matrix.
These eigenvectors are later known as Eigenfaces.
Kirby and Sirovich demonstrated that images of faces can
be linearly encoded using a modest number of basis images
. This demonstration is based on the Karhunen-Loe
transform , , , which also goes by other names,
e.g., principal component analysis , and the Hotelling
transform . The idea is arguably proposed first by
Pearson in 1901  and then by Hotelling in 1933 .
Given a collection of n by m pixel training images
represented as a vector of size m n, basis vectors spanning
an optimal subspace are determined such that the mean
square error between the projection of the training images
onto this subspace and the original images is minimized.
They call the set of optimal basis vectors eigenpictures since
these are simply the eigenvectors of the covariance matrix
computed from the vectorized face images in the training set.
Experiments with a set of 100 images show that a face image
of 91 50 pixels can be effectively encoded using only
50 eigenpictures, while retaining a reasonable likeness (i.e.,
capturing 95 percent of the variance).
Turk and Pentland applied principal component analysis
to face recognition and detection . Similar to ,
principal component analysis on a training set of face
images is performed to generate the Eigenpictures (here
called Eigenfaces) which span a subspace (called the face
space) of the image space. Images of faces are projected onto
the subspace and clustered. Similarly, nonface training
images are projected onto the same subspace and clustered.
Since images of faces do not change radically when
projected onto the face space, while the projection of
nonface images appear quite different. To detect the
presence of a face in a scene, the distance between an
image region and the face space is computed for all
locations in the image. The distance from face space is
used as a measure of “faceness,” and the result of
calculating the distance from face space is a “face map.”
A face can then be detected from the local minima of the
face map. Many works on face detection, recognition, and
feature extractions have adopted the idea of eigenvector
decomposition and clustering.
2.4.2 Distribution-Based Methods
Sung and Poggio developed a distribution-based system for
face detection ,  which demonstrated how the
distributions of image patterns from one object class can be
learned from positive and negative examples (i.e., images) of
that class. Their system consists of two components,
distribution-based models for face/nonface patterns and a
multilayer perceptron classifier. Each face and nonface
example is first normalized and processed to a 19 19 pixel
image and treated as a 361-dimensional vector or pattern.
Next, the patterns are grouped into six face and six nonface
clusters using a modified k-means algorithm, as shown in
Fig. 6. Each cluster is represented as a multidimensional
Gaussian function with a mean image and a covariance
matrix. Fig. 7 shows the distance measures in their method.
Two distance metrics are computed between an input image
pattern and the prototype clusters. The first distance
component is the normalized Mahalanobis distance between
the test pattern and the cluster centroid, measured within a
lower-dimensional subspace spanned by the cluster’s 75
largest eigenvectors. The second distance component is the
Euclidean distance between the test pattern and its projection
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 43
Fig. 6. Face and nonface clusters used by Sung and Poggio . Their
method estimates density functions for face and nonface patterns using
a set of Gaussians. The centers of these Gaussians are shown on the
right (Courtesy of K.-K. Sung and T. Poggio).
onto the 75-dimensional subspace. This distance component
accounts for pattern differences not captured by the first
distance component. The last step is to use a multilayer
perceptron (MLP) network to classify face window patterns
from nonface patterns using the twelve pairs of distances to
each face and nonface cluster. The classifier is trained using
standard backpropagation from a database of 47,316 window
patterns. There are 4,150 positive examples of face patterns
and the rest are nonface patterns. Note that it is easy to collect
a representative sample face patterns, but much more
difficult to get a representative sample of nonface patterns.
This problem is alleviated by a bootstrap method that
selectively adds images to the training set as training
progress. Starting with a small set of nonface examples in
the training set, the MLP classifier is trained with this
database of examples. Then, they run the face detector on a
sequence of random images and collect all the nonface
patterns that the current system wrongly classifies as faces.
These false positives are then added to the training database
as new nonface examples. This bootstrap method avoids the
problem of explicitly collecting a representative sample of
nonface patterns and has been used in later works , .
A probabilistic visual learning method based on density
estimation in a high-dimensional space using an eigenspace
decomposition was developed by Moghaddam and Pentland
. Principal component analysis (PCA) is used to define
the subspace best representing a set of face patterns. These
principal components preserve the major linear correlations
in the data and discard the minor ones. This method
decomposes the vector space into two mutually exclusive
and complementary subspaces: the principal subspace (or
feature space) and its orthogonal complement. Therefore, the
target density is decomposed into two components: the
density in the principal subspace (spanned by the principal
components) and its orthogonal complement (which is
discarded in standard PCA) (See Fig. 8). A multivariate
Gaussian and a mixture of Gaussians are used to learn the
statistics of the local features of a face. These probability
densities are then used for object detection based on
maximum likelihood estimation. The proposed method has
been applied to face localization, coding, and recognition.
Compared with the classic eigenface approach , the
proposed method shows better performance in face recogni-
tion. In terms of face detection, this technique has only been
demonstrated on localization; see also .
In , a detection method based on a mixture of factor
analyses was proposed. Factor analysis (FA) is a statistical
method for modeling the covariance structure of high
dimensional data using a small number of latent variables.
FA is analogous to principal component analysis (PCA) in
several aspects. However, PCA, unlike FA, does not define a
proper density model for the data since the cost of coding a
data point is equal anywhere along the principal component
subspace (i.e., the density is unnormalized along these
directions). Further, PCA is not robust to independent noise
in the features of the data since the principal components
maximize the variances of the input data, thereby retaining
unwanted variations. Synthetic and real examples in ,
, ,  have shown that the projected samples from
different classes in the PCA subspace can often be smeared.
For the cases where the samples have certain structure, PCA is
suboptimal from the classification standpoint. Hinton et al.
have applied FA to digit recognition, and they compare the
performance of PCA and FA models . A mixture model of
factor analyzers has recently been extended  and applied
to face recognition . Both studies show that FA performs
better than PCA in digit and face recognition. Since pose,
orientation, expression, and lighting affect the appearance of
a human face, the distribution of faces in the image space can
be better represented by a multimodal density model where
each modality captures certain characteristics of certain face
appearances. They present a probabilistic method that uses a
mixture of factor analyzers (MFA) to detect faces with wide
variations. The parameters in the mixture model are
estimated using an EM algorithm.
A second method in  uses Fisher’s Linear Discrimi-
nant (FLD) to project samples from the high dimensional
image space to a lower dimensional feature space. Recently,
the Fisherface method  and others ,  based on
linear discriminant analysis have been shown to outperform
the widely used Eigenface method  in face recognition on
several data sets, including the Yale face database where face
images are taken under varying lighting conditions. One
possible explanation is that FLD provides a better projection
than PCA for pattern classification since it aims to find the
most discriminant projection direction. Consequently, the
classification results in the projected subspace may be
44 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
Fig. 7. The distance measures used by Sung and Poggio . Two
distance metrics are computed between an input image pattern and the
prototype clusters. (a) Given a test pattern, the distance between that
image pattern and each cluster is computed. A set of 12 distances
between the test pattern and the model’s 12 cluster centroids. (b) Each
distance measurement between the test pattern and a cluster centroid is
a two-value distance metric. D
is a Mahalanobis distance between the
test pattern’s projection and the cluster centroid in a subspace spanned
by the cluster’s 75 largest eigenvectors. D
is the Euclidean distance
between the test pattern and its projection in the subspace. Therefore, a
distance vector of 24 values is formed for each test pattern and is used
by a multilayer perceptron to determine whether the input pattern
belongs to the face class or not (Courtesy of K.-K. Sung and T. Poggio).
Fig. 8. Decomposition of a face image space into the principal subspace
F and its orthogonal complement
F for an arbitrary density. Every data
point x is decomposed into two components: distance in feature space
(DIFS) and distance from feature space (DFFS)  (Courtesy of
B. Moghaddam and A. Pentland).
superior than other methods. (See  for a discussion about
training set size.) In the second proposed method, they
decompose the training face and nonface samples into several
subclasses using Kohonen’s Self Organizing Map (SOM) .
Fig. 9 shows a prototype of each face class. From these
relabeled samples, the within-class and between-class scatter
matrices are computed, thereby generating the optimal
projection based on FLD. For each subclass, its density is
modeled as a Gaussian whose parameters are estimated
using maximum-likelihood . To detect faces, each input
image is scanned with a rectangular window in which the
class-dependent probability is computed. The maximum-
likelihood decision rule is used to determine whether a face is
detected or not. Both methods in  have been tested using
the databases in ,  which together consist of
225 images with 619 faces, and experimental results show
that these two methods have detection rates of 92.3 percent for
MFA and 93.6 percent for the FLD-based method.
2.4.3 Neural Networks
Neural networks have been applied successfully in many
pattern recognition problems, such as optical character
recognition, object recognition, and autonomous robot driv-
ing. Since face detection can be treated as a two class pattern
recognition problem, various neural network architectures
have been proposed. The advantage of using neural networks
for face detection is the feasibility of training a system to
capture the complex class conditional density of face patterns.
However, one drawback is that the network architecture has
to be extensively tuned (number of layers, number of nodes,
learning rates, etc.) to get exceptional performance.
An early method using hierarchical neural networks was
proposed by Agui et al. . The first stage consists of two
parallel subnetworks in which the inputs are intensity values
from an original image and intensity values from filtered
image using a 3 3 Sobel filter. The inputs to the second stage
network consist of the outputs from the subnetworks and
extracted feature values such as the standard deviation of the
pixel values in the input pattern, a ratio of the number of
white pixels to the total number of binarized pixels in a
window, and geometric moments. An output value at the
second stage indicates the presence of a face in the input
region. Experimental results show that this method is able to
detect faces if all faces in the test images have the same size.
Propp and Samal developed one of the earliest neural
networks for face detection . Their network consists of
four layers with 1,024 input units, 256 units in the first hidden
layer, eight units in the second hidden layer, and two output
units. A similar hierarchical neural network is later proposed
by . The early method by Soulie et al.  scans an input
image with a time-delay neural network  (with a
receptive field of 20 25 pixels) to detect faces. To cope with
size variation, the input image is decomposed using wavelet
transforms. They reported a false negative rate of 2.7 percent
and false positive rate of 0.5 percent from a test of 120 images.
In , Vaillant et al. used convolutional neural networks to
detect faces in images. Examples of face and nonface images
of 20 20 pixels are first created. One neural network is
trained to find approximate locations of faces at some scale.
Another network is trained to determine the exact position of
faces at some scale. Given an image, areas which may contain
faces are selected as face candidates by the first network.
These candidates are verified by the second network. Burel
and Carel  proposed a neural network for face detection in
which the large number of training examples of faces and
nonfaces are compressed into fewer examples using a
Kohonen’s SOM algorithm . A multilayer perceptron is
used to learn these examples for face/background classifica-
tion. The detection phase consists of scanning each image at
various resolution. For each location and size of the scanning
window, the contents are normalized to a standard size, and
the intensity mean and variance are scaled to reduce the
effects of lighting conditions. Each normalized window is
then classified by an MLP.
Feraud and Bernier presented a detection method using
autoassociative neural networks , , . The idea is
based on  which shows an autoassociative network with
five layers is able to perform a nonlinear principal component
analysis. One autoassociative network is used to detect
frontal-view faces and another one is used to detect faces
turned up to 60 degrees to the left and right of the frontal view.
A gating network is also utilized to assign weights to frontal
and turned face detectors in an ensemble of autoassociative
networks. On a small test set of 42 images, they report a
detection rate similar to . The method has also been
employed in LISTEN  and MULTRAK .
Lin et al. presented a face detection system using
probabilistic decision-based neural network (PDBNN) .
The architecture of PDBNN is similar to a radial basis function
(RBF) network with modified learning rules and probabilistic
interpretation. Instead of converting a whole face image into a
training vector of intensity values for the neural network, they
first extract feature vectors based on intensity and edge
information in the facial region that contains eyebrows, eyes,
and nose. The extracted two feature vectors are fed into two
PDBNN’s and the fusion of the outputs determine the
classification result. Based on a set of 23 images provided by
Sung and Poggio , their experimental results show
comparable performance with the other leading neural
network-based face detectors , .
Among all the face detection methods that used neural
networks, the most significant work is arguably done by
Rowley et al. , , . A multilayer neural network
is used to learn the face and nonface patterns from face/
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 45
Fig. 9. Prototype of each face class using Kohonen’s SOM by Yang et al.
. Each prototype corresponds to the center of a cluster.
nonface images (i.e., the intensities and spatial relationships
of pixels) whereas Sung  used a neural network to find
a discriminant function to classify face and nonface patterns
using distance measures. They also used multiple neural
networks and several arbitration methods to improve
performance, while Burel and Carel  used a single
network, and Vaillant et al.  used two networks for
classification. There are two major components: multiple
neural networks (to detect face patterns) and a decision-
making module (to render the final decision from multiple
detection results). As shown in Fig. 10, the first component
of this method is a neural network that receives a
20 20 pixel region of an image and outputs a score
ranging from -1 to 1. Given a test pattern, the output of the
trained neural network indicates the evidence for a nonface
(close to -1) or face pattern (close to 1). To detect faces
anywhere in an image, the neural network is applied at all
image locations. To detect faces larger than 20 20 pixels,
the input image is repeatedly subsampled, and the network
is applied at each scale. Nearly 1,050 face samples of
various sizes, orientations, positions, and intensities are
used to train the network. In each training image, the eyes,
tip of the nose, corners, and center of the mouth are labeled
manually and used to normalize the face to the same scale,
orientation, and position. The second component of this
method is to merge overlapping detection and arbitrate
between the outputs of multiple networks. Simple arbitra-
tion schemes such as logic operators (AND/OR) and voting
are used to improve performance. Rowley et al. 
reported several systems with different arbitration schemes
that are less computationally expensive than Sung and
Poggio’s system and have higher detection rates based on a
test set of 24 images containing 144 faces.
One limitation of the methods by Rowley  and by
Sung  is that they can only detect upright, frontal faces.
Recently, Rowley et al.  extended this method to detect
rotated faces using a router network which processes each
input window to determine the possible face orientation and
then rotates the window to a canonical orientation; the rotated
window is presented to the neural networks as described
above. However, the new system has a lower detection rate on
upright faces than the upright detector. Nevertheless, the
system is able to detect 76.9 percent of faces over two large test
sets with a small number of false positives.
2.4.4 Support Vector Machines
Support Vector Machines (SVMs) were first applied to face
detection by Osuna et al. . SVMs can be considered as a
new paradigm to train polynomial function, neural networks,
or radial basis function (RBF) classifiers. While most methods
for training a classifier (e.g., Bayesian, neural networks, and
RBF) are based on of minimizing the training error, i.e.,
empirical risk, SVMs operates on another induction principle,
called structural risk minimization, which aims to minimize an
upper bound on the expected generalization error. An
SVM classifier is a linear classifier where the separating
hyperplane is chosen to minimize the expected classification
error of the unseen test patterns. This optimal hyperplane is
defined by a weighted combination of a small subset of the
training vectors, called support vectors. Estimating the
optimal hyperplane is equivalent to solving a linearly
constrained quadratic programming problem. However, the
computation is both time and memory intensive. In ,
Osuna et al. developed an efficient method to train an SVM for
large scale problems,and appliedit to face detection. Based on
two test sets of 10,000,000 test patterns of 19 19 pixels, their
system has slightly lower error rates and runs approximately
30 times faster than the system by Sung and Poggio .
SVMs have also been used to detect faces and pedestrians in
the wavelet domain , , .
2.4.5 Sparse Network of Winnows
Yang et al. proposed a method that uses SNoW learning
architecture ,  to detect faces with different features
and expressions, in different poses, and under different
lighting conditions . They also studied the effect of
learning with primitive as well as with multiscale features.
SNoW (Sparse Network of Winnows) is a sparse network of
linear functions that utilizes the Winnow update rule . It
is specifically tailored for learning in domains in which the
potential number of features taking part in decisions is very
large, but may be unknown a priori. Some of the
characteristics of this learning architecture are its sparsely
connected units, the allocation of features and links in a
data driven way, the decision mechanism, and the utiliza-
tion of an efficient update rule. In training the SNoW-based
face detector, 1,681 face images from Olivetti , UMIST
, Harvard , Yale , and FERET  databases are
used to capture the variations in face patterns. To compare
with other methods, they report results with two readily
46 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
Fig. 10. System diagram of Rowley’s method . Each face is preprocessed before feeding it to an ensemble of neural networks. Several
arbitration methods are used to determine whether a face exists based on the output of these networks (Courtesy of H. Rowley, S. Baluja, and
available data sets which contain 225 images with 619 faces
. With an error rate of 5.9 percent, this technique
performs as well as other methods evaluated on the data set
1 in , including those using neural networks ,
Kullback relative information , naive Bayes classifier
 and support vector machines , while being
computationally more efficient. See Table 4 for performance
comparisons with other face detection methods.
2.4.6 Naive Bayes Classifier
In contrast to the methods in , ,  which model
the global appearance of a face, Schneiderman and Kanade
described a naive Bayes classifier to estimate the joint
probability of local appearance and position of face patterns
(subregions of the face) at multiple resolutions . They
emphasize local appearance because some local patterns of an
object are more unique than others; the intensity patterns
around the eyes are much more distinctive than the pattern
found around the cheeks. There are two reasons for using a
naive Bayes classifier (i.e., no statistical dependency between
the subregions). First, it provides better estimation of the
conditional density functions of these subregions. Second, a
naive Bayes classifier provides a functional form of the
posterior probability to capture the joint statistics of local
appearance and position on the object. At each scale, a face
image is decomposed into four rectangular subregions. These
subregions are then projected to a lower dimensional space
using PCA and quantized into a finite set of patterns, and the
statistics of each projected subregion are estimated from the
projected samples to encode local appearance. Under this
formulation, their method decides that a face is present when
the likelihood ratio is larger than the ratio of prior
probabilities. With an error rate of 93.0 percent on data set 1
in , the proposed Bayesian approach shows comparable
performance to  and is able to detect some rotated and
profile faces. Schneiderman and Kanade later extend this
method with wavelet representations to detect profile faces
and cars .
A related method using joint statistical models of local
features was developed by Rickert et al. . Local features
are extracted by applying multiscale and multiresolution
filters to the input image. The distribution of the features
vectors (i.e., filter responses) is estimated by clustering the
data and then forming a mixture of Gaussians. After the
model is learned and further refined, test images are
classified by computing the likelihood of their feature vectors
with respect to the model. Their experimental results on face
and car detection show interesting and good results.
2.4.7 Hidden Markov Model
The underlying assumption of the Hidden Markov Model
(HMM) is that patterns can be characterized as a parametric
random process and that the parameters of this process can
be estimated in a precise, well-defined manner. In devel-
oping an HMM for a pattern recognition problem, a number
of hidden states need to be decided first to form a model.
Then, one can train HMM to learn the transitional
probability between states from the examples where each
example is represented as a sequence of observations. The
goal of training an HMM is to maximize the probability of
observing the training data by adjusting the parameters in
an HMM model with the standard Viterbi segmentation
method and Baum-Welch algorithms . After the HMM
has been trained, the output probability of an observation
determines the class to which it belongs.
Intuitively, a face pattern can be divided into several
regions such as the forehead, eyes, nose, mouth, and chin. A
face pattern can then be recognized by a process in which
these regions are observed in an appropriate order (from
top to bottom and left to right). Instead of relying on
accurate alignment as in template matching or appearance-
based methods (where facial features such as eyes and
noses need to be aligned well with respect to a reference
point), this approach aims to associate facial regions with
the states of a continuous density Hidden Markov Model.
HMM-based methods usually treat a face pattern as a
sequence of observation vectors where each vector is a strip
of pixels, as shown in Fig. 11a. During training and testing,
an image is scanned in some order (usually from top to
bottom) and an observation is taken as a block of pixels, as
shown in Fig. 11a. For face patterns, the boundaries
between strips of pixels are represented by probabilistic
transitions between states, as shown in Fig. 11b, and the
image data within a region is modeled by a multivariate
Gaussian distribution. An observation sequence consists of
all intensity values from each block. The output states
correspond to the classes to which the observations belong.
After the HMM has been trained, the output probability of
an observation determines the class to which it belongs.
HMMs have been applied to both face recognition and
localization. Samaria  showed that the states of the
HMM he trained corresponds to facial regions, as shown in
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 47
Fig. 11. Hidden Markov model for face localization. (a) Observation vectors: To train an HMM, each face sample is converted to a sequence of
observation vectors. Observation vectors are constructed from a window of W L pixels. By scanning the window vertically with P pixels of overlap,
an observation sequence is constructed. (b) Hidden states: When an HMM with five states is trained with sequences of observation vectors, the
boundaries between states are shown in (b) .
Fig. 11b. In other words, one state is responsible for
characterizing the observation vectors of human foreheads,
and another state is responsible for characterizing the
observation vectors of human eyes. For face localization, an
HMM is trained for a generic model of human faces from a
large collection of face images. If the face likelihood
obtained for each rectangular pattern in the image is above
a threshold, a face is located.
Samaria and Young applied 1D and pseudo 2D HMMs to
facial feature extraction and face recognition , .
Their HMMs exploit the structure of a face to enforce
constraints on the state transitions. Since significant facial
regions such as hair, forehead, eyes, nose, and mouth occur
in the natural order from top to bottom, each of these
regions is assigned to a state in a one-dimensional
continuous HMM. Fig. 11b shows these five hidden states.
For training, each image is uniformly segmented, from top
to bottom into five states (i.e., each image is divided into
five nonoverlapping regions of equal size). The uniform
segmentation is then replaced by the Viterbi segmentation
and the parameters in the HMM are reestimated using the
Baum-Welch algorithm. As shown in Fig. 11a, each face
image of width W and height H is divided into overlapping
blocks of height L and width W. There are P rows of
overlap between consecutive blocks in the vertical direction.
These blocks form an observation sequence for the image,
and the trained HMM is used to determine the output state.
Similar to , Nefian and Hayes applied HMMs and the
ve Transform (KLT) to face localization and
recognition . Instead of using raw intensity values, the
observation vectors consist of the (KLT) coefficients
computed from the input vectors. Their experimental
results on face recognition show a better recognition rate
than . On the MIT database, which contains 432 images
each with a single face, this pseudo 2D HMM system has a
success rate of 90 percent.
Rajagopalan et al. proposed two probabilistic methods
for face detection . In contrast to , which uses a set
of multivariate Gaussians to model the distribution of face
patterns, the first method in  uses higher order
statistics (HOS) for density estimation. Similar to ,
both the unknown distributions of faces and nonfaces are
clustered using six density functions based on higher order
statistics of the patterns. As in , a multilayer perceptron
is used for classification, and the input vector consists of
twelve distance measures (i.e., log probability) between the
image pattern and the twelve model clusters. The second
method in  uses an HMM to learn the face to nonface
and nonface to face transitions in an image. This approach
is based on generating an observation sequence from the
image and learning the HMM parameters corresponding to
this sequence. The observation sequence to be learned is
first generated by computing the distance of the subimage
to the centers of the 12 face and nonface cluster centers
estimated in the first method. After the learning completes,
the optimal state sequence is further processed for binary
classification. Experimental results show that both HOS and
HMM methods have a higher detection rate than ,
, but with more false alarms.
2.4.8 Information-Theoretical Approach
The spatial property of face pattern can be modeled through
different aspects. The contextual constraint, among others, is
a powerful one and has often been applied to texture
segmentation. The contextual constraints in a face pattern
are usually specified by a small neighborhood of pixels.
Markov random field (MRF) theory provides a convenient
and consistent way to model context-dependent entities such
as image pixels and correlated features. This is achieved by
characterizing mutual influences among such entities using
conditional MRF distributions. According to the Hammers-
ley-Clifford theorem, an MRF can be equivalently character-
ized by a Gibbs distribution and the parameters are usually
maximum a posteriori (MAP) estimates . Alternatively,
the face and nonface distributions can be estimated using
histograms. Using Kullback relative information, the Markov
process that maximizes the information-based discrimina-
tion between the two classes can be found and applied to
detection , .
Lew applied Kullback relative information  to face
detection by associating a probability function pðxÞ to the
event that the template is a face and qðxÞ to the event that the
template is not a face . A face training database consisting
of nine views of 100 individuals is used to estimate the face
distribution. The nonface probability density function is
estimated from a set of 143,000 nonface templates using
histograms. From the training sets, the most informative
pixels (MIP) are selected to maximize the Kullback relative
information between pðxÞ and qðxÞ (i.e., to give the maximum
class separation). As it turns out, the MIP distribution focuses
on the eye and mouth regions and avoids the nose area. The
MIP are then used to obtain linear features for classification
and representation using the method of Fukunaga and
Koontz . To detect faces, a window is passed over the
input image, and the distance from face space (DFFS) as
defined in  is calculated. If the DFFS to the face subspace
is lower than the distance to the nonface subspace, a face is
assumed to exist within the window.
Kullback relative information is also employed by Colme-
narez and Huang to maximize the information-based dis-
crimination between positive and negative examples of faces
. Images from the training set of each class (i.e., face and
nonface class) are analyzed as observations of a random
process and are characterized by two probability functions.
They used a family of discrete Markov processes to model the
face and background patterns and to estimate the probability
model. The learning process is converted into an optimization
problem to select the Markov process that maximizes the
information-based discrimination between the two classes.
The likelihood ratio is computed using the trained probability
model and used to detect the faces.
Qian and Huang  presented a method that
employs the strategies of both view-based and model-
based methods. First, a visual attention algorithm, which
uses high-level domain knowledge, is applied to reduce
the search space. This is achieved by selecting image areas
in which targets may appear based on the region maps
generated by a region detection algorithm (water-shed
method). Within the selected regions, faces are detected
with a combination of template matching and feature
matching methods using a hierarchical Markov random
field and maximum a posteriori estimation.
48 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
2.4.9 Inductive Learning
Inductive learning algorithms have also been applied to
locate and detect faces. Huang et al. applied Quinlan’s
C4.5 algorithm  to learn a decision tree from positive
and negative examples of face patterns . Each training
example is an 8 8 pixel window and is represented by a
vector of 30 attributes which is composed of entropy, mean,
and standard deviation of the pixel intensity values. From
these examples, C4.5 builds a classifier as a decision tree
whose leaves indicate class identity and whose nodes specify
tests to perform on a single attribute. The learned decision
tree is then used to decide whether a face exists in the input
example. The experiments show a localization accuracy rate
of 96 percent on a set of 2,340 frontal face images in the
FERET data set.
Duta and Jain  presented a method to learn the face
concept using Mitchell’s Find-S algorithm . Similar to
, they conjecture that the distribution of face patterns
pðxjfaceÞ can be approximated by a set of Gaussian clusters
and that the distance from a face instance to one of the cluster
centroids should be smaller than a fraction of the maximum
distance from the points in that cluster to its centroid. The
Find-S algorithm is then applied to learn the thresholding
distance such that faces and nonfaces can be differentiated.
This method has several distinct characteristics. First, it does
not use negative (nonface) examples, while ,  use
both positive and negative examples. Second, only the
central portion of a face is used for training. Third, feature
vectors consist of images with 32 intensity levels or textures,
while  uses full-scale intensity values as inputs. This
method achieves a detection rate of 90 percent on the first
CMU data set.
We have reviewed and classified face detection methods into
four major categories. However, some methods can be
classified into more than one category. For example, template
matching methods usually use a face model and subtem-
plates to extract facial features , , , , ,
and then use these features to locate or detect faces.
Furthermore, the boundary between knowledge-based meth-
ods and some template matching methods is blurry since the
latter usually implicitly applies human knowledge to define
the face templates , , . On the other hand, face
detection methods can also be categorized otherwise. For
example, these methods can be classified based on whether
they rely on local features , ,  or treat a face
pattern as whole (i.e., holistic) , . Nevertheless, we
think the four major classes categorize most methods
sufficiently and appropriately.
3FACE IMAGE DATABASES AND PERFORMANCE
Most face detection methods require a training data set of face
images and the databases originally developed for face
recognition experiments can be used as training sets for face
detection. Since these databases were constructed to empiri-
cally evaluate recognition algorithms in certain domains, we
first review the characteristics of these databases and their
applicability to face detection. Although numerous face
detection algorithms have been developed, most of them
have not been tested on data sets with a large number of
images. Furthermore, most experimental results are reported
using different test sets. In order to compare methods fairly, a
few benchmark data sets have recently been compiled. We
review these benchmark data sets and discuss their char-
acteristics. There are still a few issues that need to be carefully
considered in performance evaluation even when the
methods use the same test set. One issue is that researchers
have different interpretations of what a “successful detec-
tion” is. Another issue is that different training sets are used,
particularly, for appearance-based methods. We conclude
this section with a discussion of these issues.
3.1 Face Image Database
Although many face detection methods have been proposed,
less attention has been paid to the development of an image
database for face detection research. The FERET database
consists of monochrome images taken in different frontal
views and in left and right profiles . Only the upper torso
of an individual (mostly head and necks) appears in an
image on a uniform and uncluttered background. The
FERET database has been used to assess the strengthens
and weaknesses of different face recognition approaches
. Since each image consists of an individual on a uniform
and uncluttered background, it is not suitable for face
detection benchmarking. This is similar to many databases
that were created for the development and testing of face
recognition algorithms. Turk and Pentland created a face
database of 16 people  (available at ftp://whitechapel.
media.mit.edu/pub/images/). The images are taken in
frontal view with slight variability in head orientation (tilted
upright, right, and left) on a cluttered background. The face
database from AT&T Cambridge Laboratories (formerly
known as the Olivetti database) consists of 10 different
images for forty distinct subjects. (available at http://
www.uk.research.att.com/facedatabase.html) . The
images were taken at different times, varying the lighting,
facial expressions, and facial details (glasses). The Harvard
database consists of cropped, masked frontal face images
taken from a wide variety of light sources . It was used by
Hallinan for a study on face recognition under the effect
of varying illumination conditions. With 16 individuals, the
Yale face database (available at http://cvc.yale.edu/) con-
tains 10 frontal images per person, each with different facial
expressions, with and without glasses, and under different
lighting conditions . The M2VTS multimodal database
from the European ACTS projects was developed for access
control experiments using multimodal inputs . It
contains sequences of face images of 37 people. The five
sequences for each subject were taken over one week.
Each image sequence contains images from right profile
(-90 degree) to left profile (90 degree) while the subject counts
from “0” to“9” in their native languages.The UMIST database
consists of 564 images of 20 people with varying pose. The
images of each subject cover a range of poses from right
profile to frontal views . The Purdue AR database
contains over 3,276 color images of 126 people (70 males
and 56 females) in frontal view . This database is designed
for face recognition experiments under several mixing
factors, such as facial expressions, illumination conditions,
and occlusions. All the faces appear with different facial
expression (neutral, smile, anger, and scream), illumination
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 49
(left light source, right light source, and sources from both
sides), and occlusion (wearing sunglasses or scarf). The
images were taken during two sessions separated by two
weeks. All the images were taken by the same camera setup
under tightly controlled conditions of illumination and pose.
This face database has been applied to image and video
indexing as well as retrieval . Table 2 summarizes the
characteristics of the abovementioned face image databases.
3.2 Benchmark Test Sets for Face Detection
The abovementioned databases are designed mainly to
measure performance of face recognition methods and, thus,
each image contains only one individual. Therefore, such
databases can be best utilized as training sets rather than test
sets. The tacit reason for comparing classifiers on test sets is
that these data sets represent problems that systems might
face in the real world and that superior performance on these
benchmarks may translate to superior performance on other
real-world tasks. Toward this end, researchers have compiled
a wide collection of data sets from a wide variety of images.
Sung and Poggio created two databases for face detection
, . The first set consists of 301 frontal and near-
frontal mugshots of 71 different people. These images are
high quality digitized images with a fair amount of lighting
variation. The second set consists of 23 images with a total of
149 face patterns. Most of these images have complex
background with faces taking up only a small amount of the
total image area. The most widely-used face detection
database has been created by Rowley et al. , 
(available at http://www.cs.cmu.edu/~har/faces.html). It
consists of 130 images with a total of 507 frontal faces. This
data set includes 23 images of the second data set used by
Sung and Poggio . Most images contain more than one
face on a cluttered background and, so, this is a good test set to
assess algorithms which detect upright frontal faces. Fig. 12
shows some images in the data set collected by Sung and
50 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
Face Image Database
Fig. 12. Sample images in Sung and Poggio’s data set . Some images are scanned from newspapers and, thus, have low resolution. Though
most faces in the images are upright and frontal. Some faces in the images appear in different pose.
Poggio , and Fig. 13 shows images from the data set
collected by Rowley et al. .
Rowley et al. also compiled another database of images for
detecting 2D faces with frontal pose and rotation in image
plane . It contains 50 images with a total of 223 faces, of
which 210 are at angles of more than 10 degrees. Fig. 14 shows
some rotated images in this data set. To measure the
performance of detection methods on faces with profile
views, Schneiderman and Kanade gathered a set of
208 images where each image contains faces with facial
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 51
Fig. 13. Sample images in Rowley et al.’s data set . Some images contain hand-drawn cartoon faces. Most images contain more than one face
and the face size varies significantly.
Fig. 14. Sample images of Rowley et al.’s data set  which contains images with in-plane rotated faces against complex background.
expressions and in profile views . Fig. 15 shows some
images in the test set.
Recently, Kodak compiled an image database as a
common test bed for direct benchmarking of face detection
and recognition algorithms . Their database has 300
digital photos that are captured in a variety of resolutions
and face size ranges from as small as 13 13 pixels to as
large as 300 300 pixels. Table 3 summarizes the character-
istics of the abovementioned test sets for face detection.
3.3 Performance Evaluation
In order to obtain a fair empirical evaluation of face
detection methods, it is important to use a standard and
representative test set for experiments. Although many face
detection methods have been developed over the past
52 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
Fig. 15. Sample images of profile faces from Schneiderman and Kanade’s data set . This data set contains images with faces in profile views
and some with facial expressions.
Test Sets for Face Detection
decade, only a few of them have been tested on the same
data set. Table 4 summarizes the reported performance
among several appearance-based face detection methods on
two standard data sets described in the previous section.
Although Table 4 shows the performance of these
methods on the same test set, such an evaluation may not
characterize how well these methods will compare in the
field. There are a few factors that complicate the assessment
of these appearance-based methods. First, the reported
results are based on different training sets and different
tuning parameters. The number and variety of training
examples have a direct effect on the classification perfor-
mance. However, this factor is often ignored in performance
evaluation, which is an appropriate criteria if the goal is to
evaluate the systems rather than the learning methods. The
second factor is the training time and execution time.
Although the training time is usually ignored by most
systems, it may be important for real-time applications that
require online training on different data sets. Third, the
number of scanning windows in these methods vary
because they are designed to operate in different environ-
ments (i.e., to detect faces within a size range). For example,
Colmenarez and Huang argued that their method scans
more windows than others and, thus, the number of false
detections is higher than others . Furthermore, the
criteria adopted in reporting the detection rates is usually
not clearly described in most systems. Fig. 16a shows a test
image and Fig. 16b shows some subimages to be classified
as a face or nonface. Suppose that all the subimages in
Fig. 16b are classified as face patterns, some criteria may
consider all of them as “successful” detections. However, a
more strict criterion (e.g., each successful detection must
contain all the visible eyes and mouths in an image) may
classify most of them as false alarms. It is clear that a
uniform criteria should be adopted to assess different
classifiers. In , Rowley et al. adjust the criteria until the
experimental results match their intuition of what a correct
detection is, i.e., the square window should contain the eyes
and also the mouth. The criteria they eventually use is that
the center of the detected bounding box must be within four
pixels and the scale must be within a factor of 1.2 (their
scale step size) of ground truth (recorded manually).
Finally, the evaluation criteria may and should depend on
thepurpose of the detector.Ifthe detector is goingtobe used to
count people, then the sum of false positives and false
negatives is appropriate. On the other hand, if the detector is
to be used to verify that an individual is who he/she claims to
be (validation), then it may be acceptable for the face detector
tohaveadditional false detections sinceitis unlikely that these
false detections will be acceptable images of the individual,
i.e., the validation process will reject the false detections. In
other words, the penalty or cost of one type of error should be
properly weighted such that one can build an optimal
classifier using Bayes decision rule (See Sections 2.2-2.4 in
). This argument is supported by a recent study which
points out the accuracy of the classifier (i.e., detection rate in
face detection) is not an appropriate goal for many of the real-
world task . One reason is that classification accuracy
assumes equal misclassification costs. This assumption is
problematicbecause for most real-worldproblems one type of
classification error is much more expensive than another. In
some face detection applications, it is important that all the
existing faces are detected. Another reason is accuracy
maximization assumes that the class distribution is known
for the target environment. In other words, we assume the test
data sets represent the “true” working environment for the
face detectors. However, this assumption is rarely justified.
When detection methods are used within real systems, it
is important to consider what computational resources are
required, particularly, time and memory. Accuracy may
need to be sacrificed for for speed.
YANG ET AL.: DETECTING FACES IN IMAGES: A SURVEY 53
Experimental Results on Images from Test Set 1 (125 Images with 483 Faces) and
Test Set 2 (23 Images with 136 Faces) (See Text for Details)
Fig. 16. (a) Test image. (b) Detection results. Different criteria lead to
different detection results. Suppose all the subimages in (b) are
classified as face patterns by a classifier. A loose criterion may declare
all the faces as “successful” detections, while a more strict one would
declare most of them as nonfaces.
The scope of the considered techniques in evaluation is
also important. In this survey, we discuss at least four
different forms of the face detection problem:
1. Localization in which there is a single face and the
goal is to provide a suitable estimate of position;
scale to be used as input for face recognition.
2. In a cluttered monochrome scene, detect all faces.
3. In color images, detect (localize) all faces.
4. In a video sequence, detect and localize all faces.
An evaluation protocol should be carefully designed
when assessing these different detection situations. It
should be noted that there is a potential risk of using a
universal though modest sized standard test set. As
researchers develop new methods or “tweak” existing ones
to get better performance on the test set, they engage in a
subtle form of the unacceptable practice of “testing on the
training set.” As a consequence, the latest methods may
perform better against this hypothetical test set but not
actually perform better in practice. This can be obviated by
having a sufficiently large and representative universal test
set. Alternatively, methods could be evaluated on a smaller
test set if that test set is randomly chosen (generated) each
time the method is evaluated.
In summary, fair and effective performance evaluation
requires careful design of protocols, scope, and data sets.
Such issues have attracted much attention in numerous
vision problems , , , . However, perform-
ing this evaluation or trying to declare a “winner” is beyond
the scope of this survey. Instead, we hope that either a
consortium of researchers engaged in face detection or a
third party will take on this task. Until then, we hope that
when applicable, researchers will report the result of their
methods on the publicly available data sets described here.
As a first step toward this goal, we have collected sample
face detection codes and evaluation tools at http://vision.
4DISCUSSION AND CONCLUSION
This paper attempts to provide a comprehensive survey of
research on face detection and to provide some structural
categories for the methods described in over 150 papers.
When appropriate, we have reported on the relative
performance of methods. But, in so doing, we are cognizant
that there is a lack of uniformity in how methods are
evaluated and, so, it is imprudent to explicitly declare which
methods indeed have the lowest error rates. Instead, we urge
members of the community to develop and share test sets and
to report results on already available test sets. We also feel the
community needs to more seriously consider systematic
performance evaluation: This would allow users of the face
detection algorithms to know which ones are competitive in
which domains. It will also spur researchers to produce truly
more effective face detection algorithms.
Although significant progress has been made in the last
two decades, there is still work to be done, and we believe
that a robust face detection system should be effective
under full variation in:
. lighting conditions,
. orientation, pose, and partial occlusion,
. facial expression, and
. presence of glasses, facial hair, and a variety of hair
Face detection is a challenging and interesting problem in
and of itself. However, it can also be seen as a one of the few
attempts at solving one of the grand challenges of computer
vision, the recognition of object classes. The class of faces
admits a great deal of shape, color, and albedo variability due
to differences in individuals, nonrigidity, facial hair, glasses,
and makeup. Images are formed under variable lighting and
3D pose and may have cluttered backgrounds. Hence, face
detection research confronts the full range of challenges
found in general purpose, object class recognition. However,
the class of faces also has very apparent regularities that are
exploited by many heuristic or model-based methods or are
readily “learned” in data-driven methods. One expects some
regularities when defining classes in general, but they may
not be so apparent. Finally, though faces have tremendous
within-class variability, face detection remains a two class
recognition problem (face versus nonface).
The authors would like to thank Kevin Bowyer and the
anonymous reviewers for their comments and suggestions.
The authors also thank Baback Moghaddam, Henry Rowley,
Brian Scassellati, Henry Schneiderman, Kah-Kay Sung, and
Kin Choong Yow for providing images. M.-H. Yang was
supported by ONR grant N00014-00-1-009 and Ray Ozzie
Fellowship. D. J. Kriegman was supported in part by
US National Science Foundation ITR CCR 00-86094 and the
National Institute of Health R01-EY 12691-01. N. Ahuja was
supported in part by US Office of Naval Research grant
 T. Agui, Y. Kokubo, H. Nagashashi, and T. Nagao, “Extraction of
FaceRecognition from Monochromatic Photographs Using Neural
Networks,” Proc. Second Int’l Conf. Automation, Robotics, and
Computer Vision, vol. 1, pp. 18.8.1-18.8.5, 1992.
 N. Ahuja, “A Transform for Multiscale Image Segmentation by
Integrated Edge and Region Detection,” IEEE Trans. Pattern
Analysis and Machine Intelligence, vol. 18, no. 9, pp. 1211-1235,
 Y. Amit, D. Geman, and B. Jedynak, “Efficient Focusing and
Face Detection,” Face Recognition: From Theory to Applications,
H. Wechsler, P.J. Phillips, V. Bruce, F. Fogelman-Soulie, and
T.S. Huang, eds., vol. 163, pp. 124-156, 1998.
 Y. Amit, D. Geman, and K. Wilder, “Joint Induction of Shape
Features and Tree Classifiers,” IEEE Trans. Pattern Analysis and
Machine Intelligence, vol. 19, no. 11, pp. 1300-1305, Nov. 1997.
 T.W. Anderson, An Introduction to Multivariate Statistical Analysis.
New York: John Wiley, 1984.
 M.F. Augusteijn and T.L. Skujca, “Identification of Human Faces
through Texture-Based Feature Recognition and Neural Network
Technology,” Proc. IEEE Conf. Neural Networks, pp. 392-398, 1993.
 P. Belhumeur, J. Hespanha, and D. Kriegman, “Eigenfaces vs.
Fisherfaces: Recognition Using Class Specific Linear Projection,”
IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 7,
pp. 711-720, 1997.
 O. Bernier, M. Collobert, R. Fe
raud, V. Lemarie, J.E. Viallet, and D.
Collobert, “MULTRAK: A System for Automatic Multiperson
Localization and Tracking in Real-Time,” Proc. IEEE Int’l Conf.
Image Processing, pp. 136-140, 1998.
 C.M. Bishop, Neural Networks for Pattern Recognition. Oxford Univ.
54 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 1, JANUARY 2002
 C. Breazeal and B. Scassellati, “A Context-Dependent Attention
System for a Social Robot,” Proc. 16th Int’l Joint Conf. Artificial
Intelligence, vol. 2, pp. 1146-1151, 1999.
 L. Breiman, J. Friedman, R. Olshen, and C. Stone, Classification and
Regression Trees. Wadsworth, 1984.
 G. Burel and D. Carel, “Detection and Localization of Faces on
Digital Images,” Pattern Recognition Letters, vol. 15, no. 10, pp. 963-
 M.C. Burl, T.K. Leung, and P. Perona, “Face Localization via
Shape Statistics,” Proc. First Int’l Workshop Automatic Face and
Gesture Recognition, pp. 154-159, 1995.
 J. Cai, A. Goshtasby, and C. Yu, “Detecting Human Faces in Color
Images,” Proc. 1998 Int’l Workshop Multi-Media Database Manage-
ment Systems, pp. 124-131, 1998.
 J. Canny, “A Computational Approach to Edge Detection,” IEEE
Trans. Pattern Analysis and Machine Intelligence, vol. 8, no. 6,
pp. 679-698, June 1986.
 A. Carleson, C. Cumby, J. Rosen, and D. Roth, “The SNoW
Learning Architecture,” Technical Report UIUCDCS-R-99-2101,
Univ. of Illinois at Urbana-Champaign Computer Science Dept.,
 D. Chai and K.N. Ngan, “Locating Facial Region of a Head-and-
Shoulders Color Image,” Proc. Third Int’l Conf. Automatic Face and
Gesture Recognition, pp. 124-129, 1998.
 R. Chellappa, C.L. Wilson, and S. Sirohey, “Human and Machine
Recognition of Faces: A Survey,” Proc. IEEE, vol. 83, no. 5, pp. 705-
 Q. Chen, H. Wu, and M. Yachida, “Face Detection by Fuzzy
Matching,” Proc. Fifth IEEE Int’l Conf. Computer Vision, pp. 591-596,
 D. Chetverikov and A. Lerch, “Multiresolution Face Detection,”
Theoretical Foundations of Computer Vision, vol. 69, pp. 131-140,
 K. Cho, P. Meer, and J. Cabrera, “Performance Assessment
through Bootstrap,” IEEE Trans. Pattern Analysis and Machine
Intelligence, vol. 19, no. 11, pp. 1185-1198, Nov. 1997.
 R. Cipolla and A. Blake, “The Dynamic Analysis of Apparent
Contours,” Proc. Third IEEE Int’l Conf. Computer Vision, pp. 616-
 M. Collobert, R. Fe
raud, G.L. Tourneur, O. Bernier, J.E. Viallet, Y.
Mahieux, and D. Collobert, “LISTEN: A System for Locating and
Tracking Individual Speakers,” Proc. Second Int’l Conf. Automatic
Face and Gesture Recognition, pp. 283-288, 1996.
 A.J. Colmenarez and T.S. Huang, “Face Detection with Informa-
tion-Based Maximum Discrimination,” Proc. IEEE Conf. Computer
Vision and Pattern Recognition, pp. 782-787, 1997.
 T.F. Cootes and C.J. Taylor, “Locating Faces Using Statistical
Feature Detectors,” Proc. Second Int’l Conf. Automatic Face and
Gesture Recognition, pp. 204-209, 1996.
 T. Cover and J. Thomas, Elements of Information Theory. Wiley
 I. Craw, H. Ellis, and J. Lishman, “Automatic Extraction of Face
Features,” Pattern Recognition Letters, vol. 5, pp. 183-187, 1987.
 I. Craw, D. Tock, and A. Bennett, “Finding Face Features,” Proc.
Second European Conf. Computer Vision, pp. 92-96, 1992.
 J.L. Crowley and J.M. Bedrune, “Integration and Control of
Reactive Visual Processes,” Proc. Third European Conf. Computer
Vision, vol. 2, pp. 47-58, 1994.
 J.L. Crowley and F. Berard, “Multi-Modal Tracking of Faces for
Video Communications,” Proc. IEEE Conf. Computer Vision and
Pattern Recognition, pp. 640-645, 1997.
 Y. Dai and Y. Nakano, “Extraction for Facial Images from
Complex Background Using Color Information and SGLD
Matrices,” Proc. First Int’l Workshop Automatic Face and Gesture
Recognition, pp. 238-242, 1995.
 Y. Dai and Y. Nakano, “Face-Texture Model Based on SGLD and
Its Application in Face Detection in a Color Scene,” Pattern
Recognition, vol. 29, no. 6, pp. 1007-1017, 1996.
 T. Darrell, G. Gordon, M. Harville, and J. Woodfill, “Integrated
Person Tracking Using Stereo, Color, and Pattern Detection,” Int’l
J. Computer Vision, vol. 37, no. 2, pp. 175-185, 2000.
 A. Dempster, “A Generalization of Bayesian Theory,” J. Royal
Statistical Soc., vol. 30, pp. 205-247, 1978.
 G. Donato, M.S. Bartlett, J.C. Hager, P. Ekman, and T.J. Sejnowski,
“Classifying Facial Actions,” IEEE Trans. Pattern Analysis and
Machine Intelligence, vol. 21, no. 10, pp. 974-989, Oct. 2000.
 R.O. Duda and P.E. Hart, Pattern Classification and Scene Analysis.
New York: John Wiley, 1973.
 R.O. Duda, P.E. Hart, and D.G. Stork, Pattern Classification. New
York: Wiley-Intersciance, 2001.
 N. Duta and A.K. Jain, “Learning the Human Face Concept from
Black and White Pictures,” Proc. Int’l Conf. Pattern Recognition,
pp. 1365-1367, 1998.
 G.J. Edwards, C.J. Taylor, and T. Cootes, “Learning to Identify and
Track Faces in Image Sequences.” Proc. Sixth IEEE Int’l Conf.
Computer Vision, pp. 317-322, 1998.
 I.A. Essa and A. Pentland, “Facial Expression Recognition Using a
Dynamic Model and Motion Energy,” Proc. Fifth IEEE Int’l Conf.
Computer Vision, pp. 360-367, 1995.
 S. Fahlman and C. Lebiere, “The Cascade-Correlation Learning
Architecture,” Advances in Neural Information Processing Systems 2,
D.S. Touretsky, ed., pp. 524-532, 1990.
 R. Fe
raud, “PCA, Neural Networks and Estimation for Face
Detection,” Face Recognition: From Theory to Applications,
H. Wechsler, P.J. Phillips, V.