Content uploaded by Snehanshu Saha
Author content
All content in this area was uploaded by Snehanshu Saha on Jul 19, 2014
Content may be subject to copyright.
ISBN: 978-93-81693-81-0
Snow Classifier based Rapid Face Detection
and Recognition Method
Renuka Devi M N, Snehanshu Saha*, Supriya P Deshpande
*PES School of Engineering, Bangalore, India.
Renukadm@pes.edu, snehanshusaha@pes.edu, supriya.deshpande19@gmail.com,
Abstract: Face recognition has gained substantial attention over in past decades due to its increasing demand in security
applications like video surveillance and biometric surveillance. The accuracy of any face recognition system strongly
depends on the face detection system. The stronger the face detection system the better the recognition system would be.
Despite current success, there is still ongoing research in this field to make facial recognition system very rapid and
accurate.
Face detection is one of the challenging problems in image processing and computer vision. In this paper we suggest
methods for face detection and Recognition using 1)The successive mean quantization transform for feature extraction and
face detection 2)Face detection using skin region extraction 3)Face recognition using Principal Component Analysis.
Keywords: successive mean quantization transform, Principal Component Analysis, skin region extraction, Snow classifier.
1. INTRODUCTION
Human face detection plays an important role in
applications such as biometric identification, video
conferencing, intelligent human computer interface, face
image database management and face recognition. Face
detection is not perfect because it has lots of variations of
image appearance, such as pose variation, occlusion,
image orientation, illuminating condition and facial
expression
Many researchers have proposed different
approaches to address face detection problem. They are
categorized into four categories:1) Knowledge-based
methods, 2) Feature invariant approaches, 3) Template
matching methods and 4) Apperance based methods
The knowledge-based methods are mainly designed
for face localization; the difficulty in this approach is to
detect faces in different poses since it is challenging to
enumerate all possible cases. The feature invariant
approaches aim to find structural features that exist even
when the pose, viewpoint or lighting conditions vary, and
then use these to locate faces. In template matching
methods a standard face pattern (usually frontal) is
manually predefined or parameterized by a function. The
existence of a face is determined based on the correlation
values computed for the face features
Illumination and sensor variation are major issues in
visual object detection. It is convenient to transform the
raw illumination and sensor varying image so the
information only contains the main structures of the
object. The Successive Mean Quantization Transform
(SMQT) [4] can be visualized as a tradeoff between the
number of quantization levels in the analysis and the
computational part. In this paper we have used SMQT
algorithm to extract face features from the local area of
an image.
We have proposed an extension to the SNoW
classifier, the split up SNoW ,[2] for this classification
task. The split up SNoW will utilize the result from the
original SNoW classifier and create a cascade of
classifiers to perform a faster detection. It will be
analyzed that the number of splits and the number of
weak classifiers can be random within the limits of the
complete classifier. Further, a stronger classifier will
utilize all information gained from all weaker classifiers.
Face detection is a required first step in face
recognition systems. It also has several applications in
International Conference on Computer Science and Engineering
62
International Conference on Computer Science and Engineering
63
Snow Classifier Based Rapid Face Detection and Recognition Method
areas such as video coding, video conference, crowd
surveillance and human-computer interfaces.
In our paper we have proposed a frame work for
face detection and recognition using SMQT algorithm
and split up snow classifier for rapid face detection and
Princiapl component analysis for efficient face
recognition
2. RELATED WORK
There exist many reported research work related to face
feature detection and face recognition. Automatic face
detection is a complex problem in image processing and
computer vision.
Many methods have been evolved to solve this
problem such as template matching, Fisher Linear
Discriminate, Neural Networks, adaboost detectors,
support vector machine. [3][5] Success has been
achieved with each method to varying degrees and
complexities. In proposed algorithm we used upright,
frontal face images with decent resolution and under
good lighting condition. In the field of face
recognition technique the single face from test data
base is matched with single face from the training
dataset.
3. LOCAL SMQT FEATURES
The SMQT uses an approach that performs an automatic
structural breakdown of information
[
1
].
Successive Mean Quatatization Transform
X is a pixel and D(x) be set of| D(x)|=D pixels from a
local area in an image. D(x) can be vector or matrix
SMQT transformation of the local area SMQT:
D(x)ÆM(x). M(x) is new set of values.
Model of the image can be described as I(x) = gE(x) R(x)
+b
I(x)Æ Intensity image
R(x)ÆReflectance
E(x)Æiiluminance
gÆGain factor bias termÆ b
Reflectance feature should be extracted since it
contains object structure. Assume E(x) is partially
smooth. Illuminance is considered to be constant in the
chosen local area E(x)=E,
Vx € D SMQT on the local area will yield
illumination and camera –insensitive features. Entire
local patterns which consists of the same structure will
result the similar SMQT feature for certain specified
level. Number of possible patterns using local SMQT
features will be2.
4. SPLIT UP SNOW CLASSIFIER
The SNoW learning architecture is a sparse network
of linear units over a feature space of face image [1].
One of the important
properties of SNoW is the
possibility to create lookup-tables for classification
of features.
Consider a patch w of the SMQT feature M(x), then
a classifier
nonface face
(()) (())
nn
xx
xW xW
hMx hMx
θ
∈∈
=−
∑
∑
Equation can be achieved using nonface table
e and the face table
nonfac
x
h
f
ace
x
h
nonface face
xx x
hh h=−
1
and defining
threshold .
One single lookup table
for single lookup table
classification. Let the training database contain i=1,
2,….N feature patches with SMQT features. M(x) and
the corresponding classes Non-face table and face
table trained with winnow update rule initially both table
set to 0, Weight set to 1.
There are three raining parameters t
1. threshold
∝
> 2. promotion parameter
3. demotion parameter 01
β
<
<
(())
n
xW face
x
hMx
γ
∈≤
∑
and is a face then
promotion is conducted as follows
(()) (()),
face face
xx
hMx hMxxW
=
∝∀∈
International Conference on Computer Science and Engineering
64
Snow Classifier Based Rapid Face Detection and Recognition Method
If is a non-face and
(())
x
xWhMx
nnonface
γ
∈>
( ( )) ( ( )),
face face
xx
hMx hMxxW
β
=∀
(2 )
1|()|,
xxi
i
iv
vhpxW
=
=∀∈
∑
∑
∈
'W⊂
'(())
x
xWhMx
θ
∈
=∑
then demotion takes place
Procedure is repeated until no changes occur. Training of
non-face performed repeatedly and finally single table is
created.
Feature combinations for one feature, ,
i=1,2,3……..2
Then,
Let Wbe a subset chosen to contain features with
largest relevance values. Then,
can be function as weak
classifier, rejecting no faces within the training database
5. IMPLEMENTATION RESULTS OF SMQT
FACE DETECTION.
Single Face Detection
Multiface detection
Multi face detection
Web camera face detection
Figure . 1 Face detection results
6. SKIN BASED FACE DETECTION
Skin Detection
The goal of any skin detector is to distinguish between
skin and non-skin pixels in order to reduce the search
area. When constructing a process that uses skin color as
a feature for face detection, the we usually faces two
main problems. First, what color-space to choose,
International Conference on Computer Science and Engineering
65
Snow Classifier Based Rapid Face Detection and Recognition Method
second, how perfectly the skin color distribution should
be analysed Inside our algorithm, we used illumination
invariant skin detector based on chrominance a*b* from
CIE L*a*b* color space [8].
The main work of skin detection is concentrated on
building a color model that results in minimum false
detection of non-skin pixels and maximum correct
detection for skin pixels; this can be done by selecting
appropriate thresholds for each color channel that
contains the skin tones.
Fig.
3,
figure shows the flow chart of skin detection.
Fig 4 shows the results of skin based face detection.
Color Space
L*a*b color space: CIE L*a*b* (CIELAB) is the most
appropriate color space specified by the International
Commission on Illumination .It describes all the colors
visible to the human eye and was created to serve as a
device independent model to be used as a reference. The
three coordinates of CIELAB represent the lightness of
the color (L* = 0 yields black and L* = 100 indicates
diffuse white; specular white may be higher), its position
between red/magenta and green (a*, negative values
indicate green while positive values indicate magenta)
and its position between yellow and blue (b*, negative
values indicate blue and positive values indicate yellow).
The asterisk (*) after L, a and b are part of the full name.
Since the L*a*b* model is a three-dimensional model, it
can only be represented properly in a three-dimensional
space. [8] Fig2 is color space of L*a*b.
Figure. 2 Color space of L*a*b.
Unlike the RGB and CMYK color models, Lab color
is designed to approximate human vision. It aspires to
perceptual uniformity, and its L component closely
matches human perception of lightness.
Figure.3
Flow chart of skin detection.
7. IMPLEMENTATION RESULTS OF SKIN
BASED FACE DETECTION
Output 1
(a) (b) (c)
International Conference on Computer Science and Engineering
66
Snow Classifier Based Rapid Face Detection and Recognition Method
(d) (e) (f)
Output 2
(a1) (b1) (c1)
(d1) (e1) (f1)
(a)Input RGB image, (b) L*a*b colour space, (c)
Binarized image,(d) AND operation, (e) skin region, (f)
face detection
Figure. 4
skin based face detection results
8. FACE RECOGNITION USING PRINCIPAL
COMPONENT ANALYSIS
Face recognition techniques can be roughly divided into
two main categories: global approaches and feature based
techniques. In global approaches the whole image serves
as a feature vector, while in local feature approaches a
number of fiducial or control points are extracted and
used for classification.
In our paper we have presented a successful face
recognition method and it is a holistic approach based on
principal component analysis (PCA) applied on a set of
images in order to extract a set of Eigen-images, known
as Eigenfaces. Every face is modelled as a linear
combination of a small subset of these Eigen faces and
the weights of this representation are used for
recognition. The identification of a test image is done by
locating the image in the database, whose weights are the
closest to the weights of the test image [9].
The Eigen face recognition algorithm is executed as
follows:
1. Calculate Eigen face components of new photo,
2. Calculate a difference between mean-face and the
new face,
3. Multiply difference with each eigenvector (weight),
4. Build a new vector of weights,
5. Choose a best matched face on a basis of Euclidian
distance of weights,
6. Determine whether recognized face belongs to
known faces class (by comparing the Similarity
measure with two thresholds).
9. THE PRINCIPAL COMPONENT ANALYSIS-
(EIGENFACE ALGORITHM)
The presented algorithm can be divided into two parts –
database building and image recognition [9]
The eigenface database building algorithm consists
of following steps:
• Prepare the reference set – all images should have the
same pixel resolution and the same alignment of a
face, Let the training set of face images be 1
Γ
,2
Γ
,
3
Γ
, …, M
Γ
.
• Transform images into vectors and building a matrix
from these vectors, where each the matrix row
represents a single image, Calculate a mean image
(average face),
The average face of the set if defined by
∑
=
Γ=Ψ
M
nn
M1
1. (1)
• Subtract the mean image from each image in the
matrix (the mean of the matrix rows will be equal 0),
Each face differs from the average by the vector
Ψ−
Γ
=
Φ
nn . (2)
• Calculate eigenvectors and eigenvalues of covariance
matrix.. The eigenvectors of this covariance matrix
are so-called eigenfaces, Covariance matrix
International Conference on Computer Science and Engineering
67
Snow Classifier Based Rapid Face Detection and Recognition Method
∑
=
=ΦΦ=
M
n
TT
nn AA
M
C
1
1 (3)
Where the matrix . The matrix C,
however, is by , and determining the
eigenvectors and eigenvalues is an intractable task for
typical image sizes.
]...[ 21 M
AΦΦΦ=
2
N2
N2
N
n
• Find a set of orthonormal vectors, which best
describes distribution of data
• Choose a subset from these orthonormal vectors
which are associated with highest eigenvalues of
covariance matrix,
• Sort these eigenvectors in ascending order – they will
be called as eigenfaces.
Consider the eigenvectors
ν
of such that AAT
nnn
TAA
νλν
=
nnn
TAAAA
νλν
=
n
A
(4)
Premultiplying both sides by A, we have
(5)
From which we see that
ν
are the eigenvectors of
.
T
AAC =
AAL T
=n
T
mmn
LΦΦ=
n
• Following this analysis, we construct the M by M
matrix , where , and find
the M eigenvectors
ν
of L. These vectors determine
linear combinations of the M training set face images
to form the eigenfaces n
μ
:
MnA n
M
kknkn ,......,1,
1
==Φ= ∑
=
ννμ
(6)
With this analysis the calculations are greatly
reduced, from the order of the number of pixels in the
images ( ) to the order of the number of images in the
training set (M). In practice, the training set of face
images will be relatively small ( ), and the
calculations become quite manageable. The associated
eigenvalues allow us to rank the eigenvectors according
to their usefulness in characterizing the variation among
the images [12]
2
N
2
NM <
10. RESULTS OF ONLINE FACE RECOGNITION
USING EIGEN FACE METHOD
(i)Test image (ii) Matched image
minimum Euclidean Distance value=43.12
(i)Test image (ii)Matched image
Figure 5: Results of online face recognition using Eigen
face Method
11. RESULT AND ANALYSIS
B. Similarity measure
An image similarity measure signifies the degree of
similarity between intensity patterns in two images. The
image similarity measure depends on the modality of the
images to be matched.
In our implementation we have
used similarity measure to match the face images in
principal component analysis algorithm.
.
cos ||||
AB
AB
θ
= Where A and B are values of the
Eigen vector.
International Conference on Computer Science and Engineering
68
Snow Classifier Based Rapid Face Detection and Recognition Method
Table 1
.
A
.
Euclidean Measure
Euclidean distance values for 20 Train data set
face images
43.12 minimum Euclidean value is the
→
77.99 matched value for the face(a)
214.41
80.46
110.95
317.10
312.18
278.79
184.98
184.98
132.76
176.81
90.15
133.92
112.76
113.55
148.06
82.08
708.37
101.76
Table 2. Similarity Measure
calulated theta( θ) value
θ = 0
θ = 0.81
θ= 1.75
θ=2.56
θ= 3.43
θ= 4.23
θ= 5.98
θ= 5.05
Calculated
θ
values are very close to 0-Æthe image
vectors are aligned in such a way that angle between
those is practically negligible which in turn leads to the
conclusion that both the train and test face images match
close to 100%.
12. ACKNOWLEDGEMENT
Our, sincere thanks to Dr Suryaprasad J, Director &
Principal, Dr.Srikanta Murthy, HOD, Department of
Computer Science & Engineering and Prof. Sandesh B
J,HOD, Department of Information Science and
Engineering, PES School of Engineering, Bangalore, for
their constant support and encouragement.
13. CONCLUSIONS
In this paper we have developed an efficient Face
detection and Recognition system. Successive Mean
Quantization Transform algorithm is used as feature
extraction method for object (face) detection and is
applied to locate the face in still images and also live
from web camera. We have presented another method for
model-free, nonuser- Initiated face detection based on
skin color. Based on L*a*b and YCbCr color space, skin
clusters, skin-like regions are detected in the image. we
have also implemented face recognition system using
Principal Component Analysis i.e. Eigen faces method
for efficient recognition of unknown face with a known
set of face images in the database.
REFERENCES
[1]. Mikael Nilsson, Jorgen Nordberg, and Ingvar
Claesson “Face Detection Using Local Smqt features
And Split Up Snow classifier”, Blekinge Institute of
Technology, sweden
[2]. D. Roth, M. Yang, and N. Ahuja, “A snow-based
face detector “, in In Advances in Neural Information
Processing Systems 12 (NIPS12), pp.855ñ861,
MITPress 2000.
[3]. P. Viola and M. Jones,” Rapid object detection using
a boosted cascade of simple features”, In
Proceedings of the 2001 IEEE Computer Society
Conference on Computer Vision and Pattern
Recognition (CVPR),2001,vol.1,pp.511-518.
[4]. M. Nilsson , M. Dahl, and I. Claesson , “The
successive mean Quantization transform “, in IEEE
International Conference on Acoustics, Speech, and
Signal Processing(ICASSP),March
2005,vol.4,pp.429-432.
[5]. H. Rowley, S. Baluja, and T. Kanade,”Neural
network–based Face detection,” in In Proceedings of
Computer Vision and Pattern Recognition, June
1996, pp.203-208.
[6]. Yuanyuan Liu, Haibin Yu Zhiwei He, Xueyi Ye
“Fast Robust Face Detection under a Skin Color
Model with Geometry Constraints “, 2009
International Conference on Computational
Intelligence and Security
[7]. Sanjay Kr. Singh , D. S. Chauhan , Mayank Vatsa ,
Richa Singh ,” A “Robust Skin Color Based Face
Detection Algorithm”, Tamkang Journal of Science
and Engineering, Vol. 6, No. 4, pp. 227-234 (2003)
[8]. Safwan R. Wshah and Ibrahim M. Mansour, “A
Robust Algorithm for Face Detection in Color
Images Based on Color Segmentation and Neural
Network Techniques”, Dirasat, Engineering
Sciences, Volume 33, No. 2, 2006
International Conference on Computer Science and Engineering
69
Snow Classifier Based Rapid Face Detection and Recognition Method
[9]. Vinay Hiremath & Ashwini Mayakar “Face
Recognition Using Eigen face Approach”, IDT
Workshop on Interesting Results in Computer
Science and Engineering, 2009
[10]. Parvinder S. Sandhu, Iqbaldeep Kaur, Amit
Verma, Samriti Jindal, Inderpreet Kaur, Shilpi
Kumari, “Face Recognition Using Eigen face
Coefficients and Principal Component Analysis”,
International Journal Of Electrical And Electronics
Engineering 3:8, 2009.
[11]. Parvinder S. Sandhu, Iqbaldeep Kaur, Amit
Verma, Samriti Jindal, Inderpreet Kaur, Shilpi
Kumari, “Face Recognition Using Eigen face
Coefficients and Principal Component Analysis”,
International Journal Of Electrical And Electronics
Engineering 3:8, 2009.
[12]. Srinivasulu Asadi Dr.Ch.D.V.Subba Rao
V.Saikrishna, “A Comparative study of Face
Recognition with Principal Component Analysis
and Cross-Correlation Technique”, International
Journal of Computer Applications (0975 8887)
Volume 10 No.8, November 2010
