Single frontal face detection by finding dark pixel group and comparing xy

International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME
471
SINGLE FRONTAL FACE DETECTION BY FINDING DARK PIXEL
GROUP AND COMPARING XY-VALUE OF FACIAL FEATURES
A.HEMLATA1
, MAHESH MOTWANI2
1
( Computer Science and Engg. Dept, University Institute Of Technology, Rajeev Gandhi
Proudyogiki Vishwavidyalaya, Bhopal, India)
2
( Computer Science and Engg. Dept, University Institute Of Technology, Rajeev Gandhi
Proudyogiki Vishwavidyalaya, Bhopal, India)
ABSTRACT
Single frontal face detection is important step towards multiple faces and side
faces. In this paper, a detection algorithm for single frontal face detection in colour
images is proposed. Algorithm first detects skin colour region and then skin region is
further processed by finding connected components and holes. Each connected
components is tested to extract facial features like eyes, nose and mouth and by
drawing lines between facial features and comparing the drawn triangles with each
other. We conclude by facial feature and triangle properties that the region is face. The
detected face region is enclosed by drawing rectangle using the parameters of triangle.
Keywords: connected components, holes, normalization, open area, skin color.
1. INTRODUCTION
The computing technology has improved the real-time vision modules that interact
with humans in recent years. It is first step to locate faces in a scene before any recognition
algorithm can be applied in biometric system. Face detection is an important step for the
success of any face processing system. The face detection problem is very difficult as it needs
to account for all possible appearance variation caused by change in lighting effect, facial
features, expressions, etc. In addition, it has to detect faces that appear at different scale,
position, angle of rotations. In spite of all these difficulties, great progress has been made
and many systems have shown real-time performance.
Human face detection is useful in applications such as surveillance of video, human
computer interface, recognition of face and management of image database. Detecting faces
is a prerequisite for face recognition and facial expression analysis.
INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING
& TECHNOLOGY (IJCET)
ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)
Volume 4, Issue 2, March – April (2013), pp. 471-481
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
www.jifactor.com
IJCET
© I A E M E

472
Anthropometry features can be used to detect facial features. Anthropometry is a
biological science that deals with the measurement of the human body and its different parts.
The task of human face detection is to determine in an arbitrary image whether or not
there are any human faces in the image, and if present, localize each face and its extent in the
image, regardless of its three dimensional position and orientation. Such a problem is a very
challenging task because faces are non-rigid forms and have a high degree of variability in
size, color, shape and texture.
1.1 Background
Face detection and facial feature extraction using color snake [1] is composed of three
main parts: the face region estimation part, the detection part and the facial feature extraction
part. In the face region estimation part, images are segmented based on human skin color. In
the face detection part the template matching method is used. Color snake is then applied to
extract the facial feature points within the estimated face region.
In Face Detection Based on a New Color Space YCgCr [2], a new color space YCgCr
is described and applied for face detection. The required decision values, which determine
the decision thresholds are modeled and represented in the Cg-Cr plane. Based on the
representation of the experimental results, the decision thresholds are modified. The
segmentation results achieved with this new color space are compared with those obtained in
YCbCr.
Domain Specific View for Face Perception [3] performs a skin color analysis of the
image and random measurements are generated to populate the entrant face region according
to the face anthropometrics using the eye location determined from the YCbCr color space.
Face Detection using Rectangle Features and SVM [4] is based on the characteristics
of intensity and symmetry in eye region. Three rectangle features are developed to measure
the intensity relations and symmetry. The found eye-pair-like regions and SVM are used to
find the face candidates.
Automatic Facial Features Localization [5] determines the facial features
automatically, without the participation of a human. This method uses the Active Shape
Models and Evolution Strategies, to find the facial features in images of faces in frontal view
automatically
A simple and efficient eye detection method in color images [6] detects face regions
in the image using a skin color model in the normalized RGB color space. Then, eye
candidates are extracted within these face regions. Finally, using the anthropological
characteristics of human eyes, the pairs of eye regions are selected.
1.2 Our Approach
The proposed method is based on finding xy coordinate value of pixel and relation
between facial features for the detection of face. The main steps of the approach are as
follows, and are described in their corresponding subsections in greater detail:
1. Noise Removal.
2. Skin color detection.
3. Find connected components and test each connected component.
4. Rejecting the region, having less than two holes.
5. Each accepted region is tested for facial feature.
6. Feature extraction like eyes, nose and mouth by performing the following test:
i. Hole location test to find the xy coordinates of the holes.

473
ii. Circularity test to find circle shaped holes.
iii. Features are extracted by comparing xy coordinate value of one hole with that
of other holes.
7. By drawing the lines between facial features, various triangles are formed. Compare
the triangles to conclude the region as face and also use the triangle to draw rectangle
enclosing the face.
2. METHOD
2.1 Noise Removal
Median filter is used to remove noise from image. It examines the neighborhood
pixel values, sort the pixels and replace the current pixel value by the median value.
2.2 Skin Color Detection
Image is segmented by separating human skin regions from non-skin regions based on
color, a reliable skin color model [7] that is adaptable to people of different skin colors and to
different lighting conditions is used. The common RGB representation of color images is not
suitable for characterizing skin-color. In the RGB space, the triple component (r, g, b)
represents color and luminance. Luminance varies in faces due to the ambient lighting and
is not a reliable measure in separating skin from non-skin region. Luminance is removed
from the color representation by using chromatic color space [8]. Chromatic colors are also
known as "pure" colors is obtained by normalization process. The chromaticity r, g and b are
defined as:
r = R/(R+G+B)
b = B/(R+G+B)
We use only r and b to describe the skin color, given that the third component depends
on the other two (g = 1- r -b).
Skin color model is created using 11 different images of varying colors to create skin model.
Crop small portion of skin from each image. Convert the colors into chromatic color space to
eliminate luminance. Each sample was low pass filtered to remove noise. The low pass filter
is determined by:
1 1 1
1/9 1 1 1
1 1 1
The sample images are converted from RGB to YCbCr space and luminance Y
component is discarded. Then the sample means for both Cb and Cr components and the
covariance matrix are computed in order to fit the data into Gaussian Model. The Normal
distribution is also known as the Gaussian Distribution and the curve is also known as the
Gaussian Curve, named after German Mathematician-Astronomer Carl Frederich Gauss. The
probable skin regions in original image using this distribution is obtained by:

474
P(r,b) = exp[-0.5(x-m)T
C-1
(x-m)]
Where x =(r,b)T
and T denotes matrix transpose, C= covariance, m = mean .
This algorithm determines the probability of one pixel being a skin region based on the skin
region model. Create a grayscale image giving the probability based on the gray level .
Fig 1. Skin Color Image
2.3 Finding Connected Components And Holes
In image processing, connected components labeling [9,10] scans an image and groups its
pixels based on connectivity feature, i.e. all pixels in a connected component share similar pixel
intensity values and are in some way connected with each other. Find all such connected
component and then label each pixel with a gray level or a color (color labeling) according to
the component it was assigned to. A connected components means finding regions of connected
pixels which have the same value. The connected components may be face or other skin area of
human like hand, neck etc. Each connected component is tested to find whether the region
contains face or not.
Now we find number of holes [9] in each connected component to detect whether the
connected component contains face or not. Hole is an area of dark pixels surrounded by lighter
pixels is shown in Fig 2(a). Dark pixels are black color pixel and lighter pixel means white color
pixel. If the connected component contains more than one holes then the region may contain
face, is accepted for further processing otherwise the region is rejected.
2.4 Rectangular Box
Find the size of each connected component and enclose it by rectangle and also store the
minimum xy-value and maximum xy-value to process the rectangle further.
2.5 Facial Feature Extraction
Each bounding box image has to undergo the following test individually for feature
extraction.
i. Hole Location test: The location of holes in the bounding box image can be detected by
finding connected component in the negative image of bounding box.
Fig 2(a). Connected component Fig 2(b). Bounding box Fig 2(c). Negative
and Holes of Connected component image of Connected
component
Hole : Dark
(black) pixel
Lighter (White)
pixel group

475
ii. Circularity test: The circularity test is performed on each internal holes i.e.
Connected component in negative image of the bounding box image.
Circularity= Perimeter^2 / (4 * pi *Area)
If the circularity value is in the range from 0.5 to 1.2 , we can conclude that
this is closer to circle shape. The holes concluded as circle has to undergo further
testing for feature extraction.
Then each bounding box is to be tested by using the minimum and maximum
xy- value of rectangular box to find facial features [11] like eyes, nose and mouth
etc.
• Eye feature extraction
(i) Find first hole (black pixel group) within Fig 2(c) confirmed as circle. Reject
the holes which is not confirmed as circle. If it is first eye as shown in Fig 3(a), then
find xeye11 minimum x-value of first eye, yeye11 is minimum y-value of first eye.
xeye12 is maximum x-value of first eye, yeye12 is maximum y-value of first eye.
(ii) Find next hole (black pixel group) confirmed as circle that lies in the same range
of yeye11 to yeye12 and whose x-value is greater than x-value of first eye as shown in
Fig 3(b). Here it is assumed that if these black pixel groups are eyes then both black
pixel groups should be in horizontal line means almost same y value or some pixel’s
y-values of one group matches with the other group. To confirm these two holes as
eyes, find the area and perimeter of each and check the condition:
Perimetereye1 - Perimetereye2 < 5
Areaeye1 - Areaeye2 < 12
The two eyes are almost equal in area and perimeter. Hence in above condition, the
difference of perimeter of two eyes should be less than 5 and the difference of area of
two eyes should be less than 12 are concluded by performing sample test of two
eyes of various images. If the above condition becomes true, then we can conclude
that it is likely to be eyes.
xeye21 is minimum x-value of second eye and yeye21 is minimum y-value of
second eye. xeye22 is maximum x-value of second eye, yeye22 is maximum y-value of
second eye. Then check for the given case conditions:
xeye11<xeye21, xeye11<xeye22
xeye12<xeye21, xeye12<xeye22
yeye11=yeye12, yeye21
(There may be more than one y-value in which at least one pixel value of the group
should satisfy this same y-value condition)
Fig 3(a). Box with Fig 3(b). Box with Fig 3(c). Box with
one eye two eyes two eyes and nose

476
Once both eyes are located , we try to find nose and mouth pixel group. If both
eye is not in the image , then also we try to find nose and mouth pixel group.
• Nose and mouth feature extraction
Case I: If both eyes are found in previous step as shown in Fig 6(a).
By using the maximum x-value of first eye xeye12 and minimum x-value of second
eye xeye21, we can find nose and mouth features.
For Nose, find the next black pixel group with y-values higher than that of found eyes
and x-value lies between xeye12 and xeye21 as shown in Fig 3(c).
xeye12<=xn<= xeye21
yeye12<yn and yeye22<yn
Where (xn,yn) is xy-value of nose.
For Mouth , suppose (xm,ym) is xy-value of mouth.
xeye11<xm<xeye22
yeye12< ym and yeye22< ym
The x-value of this pixel group is almost equal to the Eyes x-value range and y-value
is greater than eye’s y-value. We can conclude that this group is either nose or mouth.
Store the x -value and y-value of this black pixel group . Then find another
group , if such group exists whose x-value lies in the range of eyes x-value range.
Apply case I :test for this dark pixel group. If the result of this test is true then this
pixel group may be nose or mouth. To exactly find nose and mouth , compare the y
value of both pixel group, lesser y-value group may be nose and group having higher
y-value may be mouth.
yn<ym
Case II : If both eyes are not in the image as shown in Fig 6(b).
We try to find two black pixel group that lies in vertically and then test for nose and
mouth.
Fig 6(a). Box with two eyes, Fig 6(b). Box with one eye,
nose and mouth nose and mouth
Case III: If nose does not exist in the image as shown in Fig 7(a)., then mouth’s x-
value lies in the range of eyes x-value.
Case IV: If mouth does not exist in the image as shown in Fig 7(b)., then nose’s x-
value lies in the range of eyes x-value.

477
Fig 7(a). Box with two Fig 7(b). Box with two Fig 7(c). Box with
eyes and mouth eyes and nose two eyes, nose and mouth
i. If both mouth and nose is not in the image then also by finding eye’s location we can
say that it might be a face but not sure.
ii. If eyes, nose and mouth exists as shown in Fig 7(c)., then we can conclude that the
region is face. Draw the rectangle to enclose exactly the face region by using the
coordinate values of eyes, nose and mouth.
Draw line between two eyes AC. Draw line BF and FI passing through nose
and mouth, perpendicular to AC, B is midpoint of AC, I is mouth region. Draw line
DH which is almost 2.5 to 3 times of AC in length, F is midpoint of DH. The line
between eyes, nose and mouth are drawn in such a way that the following conditions
exists:-
a. Triangle2 is congruent to Triangle3, Triangle4 and Triangle5.
b. Triangle1 is congruent to Triangle6.
c. Triangle7 is congruent to Triangle8.
If the above conditions are true then we can confirm that the tested black pixel
group A and C are eyes, I is mouth, nose is in the line BF. D and H are the
approximate location of ears. Now we have to draw the rectangle enclosing the face.
If the length of AC is X then it is concluded by testing many sample faces that
the length of LM is nearly 2.5X to 3X and length JL is nearly 3.6X. Draw JL 3.6
times of AC in length and complete the rectangle.
Fig 8. Triangular figure connecting facial features

478
Fig 9. Face with Triangular figure
EXPERIMENTAL RESULTS
In this paper, the experiments are performed in MATLAB to verify the effectiveness
of the method. In order to verify the validity, 20 images are tested. The experiment shows
reasonably good result using the proposed method for locating face on image containing
frontal face. But false detections are still existing .The statistical data is shown in Table.1
Table 1. Statistical data
Face Numbers In A Photo 1
Photo Numbers 20
Face Numbers 20
Hits 20
Wrong Selected 1
False Detection Rate 5%
Missed Rate 0%
Precision Rate 100%

479
(a) (b) (c)
(d) (e)
Fig 10 (a) shows input image, (b) shows detected skin Region, (c) shows binary image
after skin region detection, (d) shows hole detected image, (e) shows face with triangular
figure
(a) (b)
(c) (d)
Fig 11 (a) shows input image, (b) shows binary image after skin region detection, (c)
shows hole detected image, (d) shows face with triangular figure.

480
CONSTRAINTS IN PROPOSED METHOD
The Frontal Face with no orientation is detected with high accuracy. If the face is
tilted by some angle, it is important to pose the head straight. Because, the above
algorithm are trying to find first eye and it is assumed that next eye will lie horizontally
having nearly same y-value. If head’s position is not straight then it is difficult to find the
next eye and other features with this algorithm.
CONCLUSIONS
We have presented a method for face detection using feature extraction model.
This detection method initially takes the color image and by combining Gaussian skin-
color model, detects skin color region. Then by finding connected components, feature is
extracted from each connected component, face is detected. This method works with the
image containing vertically straight head. Experimental results show high detection
accuracy. In the future work, the algorithm can be improved to detect tilted face, multiple
face and side face to achieve better performance and further reduce the false detecting
rate in dealing with images with more complex background.
REFERENCES
[1] Kap-ho seo, Won kim, Changmok oh, Ju-Jang Lee, “Face detection and facial
feature extraction using color snake,” IEEE 2002.
[2] Juan José de Dios Narciso García, “FACE DETECTION BASED ON A NEW
COLOR SPACE YCgCr,” IEEE 2003.
[3] Mary Metilda and T. Santhanam, “Domain Specific View for Face Perception”,
Information Technology Journal 7(1), pp 105-111, 2008.
[4] Qiong Wang, Jingyu Yang, and Wankou Yang, “Face Detection using Rectangle
Features and SVM,” International Journal of Intelligent Systems and Technologies,
2006.
[5] Vittorio Zanella, Héctor Vargas, Lorna V. Rosas, “Automatic Facial Feature
Localization”, IEEE 2005.
[6] D. Sidibe, P. Montesinos, S. Janaqi, “A simple and efficient eye detection method in
color images”, International Conference Image and Vision Computing, New
Zealand, 2006.
[7] Quan Huynh-Thu, Mitsuhiko Meguro, Masahide Kaneko, “Skin-Color-Based
Image Segmentation and Its Application in Face Detection” MVA2002 IAPR
Workshop on Machine Vision Applications 2002, Japan.
[8] Tarek M. Mahmoud, “A New Fast Skin Color Detection Technique,” World
Academy of Science, Engineering and Technology, 43, 2008.
[9] A. Rakhmadi, M. S. M. Rahim, A. Bade, H. Haron, I. M. Amin, “Loop Back
Connected Component Labeling Algorithm and Its Implementation in Detecting
Face”, World Academy of Science, Engineering and Technology 40, 2010.

481
[10] Jean V. Fonou Dombeu and Jules-R. Tapamo, “Validation of Detected Facial
Components For an Accurate Face Recognition”, Proceedings of Pattern Recognition
Association of South Africa( PRASA), 2007.
[11] Cooray, Saman H. and O'Connor, Noel E. , “Facial feature extraction and principal
component analysis for face detection in color images”, International Conference on
Image Analysis and Recognition ICIAR, 2004, Porto, Portugal.
[12] Jyoti Verma and Vineet Richariya, “Face Detection and Recognition Model Based on
Skin Colour and Edge Information for Frontal Face Images”, International journal of
Computer Engineering & Technology (IJCET), Volume 3, Issue 3, 2012, pp. 384 - 393,
ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375.
[13] Mr. K. Gnanamuthu Prakash and Dr. S. Balasubramanian, “Literature Review of Facial
Modeling and Animation Techniques”, International Journal of Graphics and
Multimedia (IJGM), Volume 1, Issue 1, 2010, pp. 1 - 14, ISSN Print: 0976 – 6448,
ISSN Online: 0976 –6456.
[14] Sambhunath Biswas and Amrita Biswas, “Fourier Mellin Transform Based Face
Recognition”, International journal of Computer Engineering & Technology (IJCET),
Volume 4, Issue 1, 2013, pp. 8 - 15, ISSN Print: 0976 – 6367, ISSN Online:
0976 – 6375.
AUTHORS’ DETAIL
Mahesh Motwani is an Associate Professor (CSE) at University
Institute of Technology, RGPV, and Bhopal. Presently, he is acting as
Secretary of Polytechnique wing, RGPV, Bhopal. He has worked as a
Deputy Registrar (Examination) in Rajiv Gandhi Technological
University, Bhopal and also as Associate Professor (CSE) in Govt.
Engineering College, Jabalpur. He has teaching experience of about 19
years. He received his master degree from the Indian Institute of
Technology (IIT) Delhi and Ph.D. from Rajiv Gandhi Technological
University, Bhopal in computer science and engineering. His research areas includes
Computer vision, Data mining, Image processing, Wireless and Adhoc Networks, in which he
has published about 30 scientific papers. He is the author of a book on Cryptography &
Network security.
A.Hemlata has received B.E and M.Tech in computer science and
engineering, is currently pursuing Ph.D. from the Rajiv Gandhi
Technological University, Bhopal. She has worked as Assistant
Professor (CSE) in Govt. Engineering College, Jabalpur and currently
working as Assistant Professor (CSE) in University institute of
technology, Rajiv Gandhi Technological University, Bhopal. She is
member of Institution of Engineers, IAENG and IACSIT. Her research
area includes Computer Vision and Image processing.

Single frontal face detection by finding dark pixel group and comparing xy

More Related Content

What's hot (19)

Viewers also liked (9)

Similar to Single frontal face detection by finding dark pixel group and comparing xy (20)

More from IAEME Publication (20)

Recently uploaded (20)

Single frontal face detection by finding dark pixel group and comparing xy