Robust technology based on appearance -methods to [602596]

Robust technology based on appearance -methods to
reconstruct a couple of image s

B. Satouri
LIIAN, Department of Computer sciences
Faculty of Sciences Dhar -Mahraz P.O.Box 1796 Atlas –
Fes, Morocco
[anonimizat]

K. Satori
LIIAN, Department of Computer sciences
Faculty of Sciences Dhar -Mahraz P.O.Box 1796 Atlas –
Fes, Morocco
khalidsatori@gmail .fr

A. El abderrahmani
LIIAN, Department of Mathematics and informatics
Larache Poly disciplinary School,
LARACHE, Moro cco
[anonimizat]

Abstract— In this paper, we present a new and important
meth od that will be used in the match ing, self -calibration
and 3D reconstructi on of three -dimensional scenes. The
particularity of our approach resid es in the minimization
of the number of image in input i.e. The use of a single
image instead of a couple to provide data neede d for other
domains. Our new approach is based primarily on
detecting the face in the input image as well as detecting
the eyes in the face detected using the appearance methods
to provide a couple of image in output. The algorithms
used and the practical results obtained show the
importance and the performance of the proposed
approach

Keywords — Detection of face , Detection of eyes, Appearance
methods

I. INTRODUCTION
Computer vision is the science of vision machines. It is a
scientist discipline who is interested in building artificial
systems that obtain information from images. The input data
can take many forms: photographs, video footage, multiple
camera images or multidimensional data medical scanner.
Subdomains of computer vision are for example the
Reconstruction of scenes, detection of events, object
recognition, learning and image restoration.
Detecting the face in the image is an essential and crucial
treatme nt before the reconstruction phase. In fact, the process
of reconstruct ing faces can never become fully automatic if it
has not been preceded by an efficient detection. The
processing consists in searching in an image Position of the
faces and extracts the m in the form of a set of thumbnails in
the purpose of facilitating their further processing. A face is
considered correctly detected if the extracted thumbnail size
does not exceed 20% of the actual size of the Facial area, and essentially contains the ey es, nose and mouth
[1, 2]. Recently, neural networks [24 -26], support vector
machines [27 ], kernel methods [28], and ensemble techniques
[27] also find great applications in this area.
The detection of the eyes [29 -31] in image is an
indispensable and cruc ial before the recognition phase of iris
that learns in the field of biomedicine. In thi s way, the
recognition process of iris can never become fully automatic if
it has not been preceded by an eye -detecting step that comes
after the face detection phase. The treaty consists in seeking an
image the position of the two eyes and extract s them in the
form of a set of thumbnails for the purpose of facilitating their
processing at a later date.

Our approach is a new method for the construction of a
couple of images using a single human image based on two
steps: face detection and eye detection that will be used in the
matching, self -calibration and 3D reconstruction of three –
dimensional scenes.

Our work is organized as follows: In the third part, we
present the stages of implementat ion of our method (Figure 1 )
containing two subparts: the first subpart comprises the
detection of face which is based on appearance -methods,
precisely the Viola -Jones and skin color pixel method .
Finally, a relevant interpretation and experimentation will
be provided in the fourth part.

II. SURVEY OF THE PREVIOUS WORKS

A classification of the methods of facial localization is
proposed by Yang Et al. [4]. The methods are divided into
four categories. These categories can be Overlap if an
algorithm can belong to two or more categories. This
classification may be made as follows:

1) APPROACH BASED ON RECOGNITION
These methods are based on the knowledge of the various
elements that kill a face and relationships that exist between
them. Thus, the relative positions of various key elements such
as the mouth, nose and eyes are measured to and then to the
'no visual ' face classification in Chiang et al. [5]. The problem
in this type of method is that it is difficult to define well in a
single way a face. If the definition is too detailed, some faces
will be missed while if the description is too general, the rate
of false positives will rise.

2) APPROACH BASED ON INVARIANT
CHARACTERISTICS
The goal of this algorithm family is to find the structural
Even if the face is in different positions, light conditions or
Angle of view. The problem with this approach is that the
quality of the images May be s everely diminished due to
illumin ation, noise or occlusion. However, there are several invariable properties or characteristics of the face of which the
main ones are as follows:

Skin color: The skin color of the human being has been used
to the detection of faces and its relevance has been proven as a
feature Specific to the face. The principle of this method is
based on information Color for discrimination of skin or non –
skin pixels. Each pixel of a color image is encoded in a color
space

3) APPROACH BASED ON THE MATCHING OF
TEMPLATES
The detection of whole faces or parts of the face is done
through a learning of standard examples of faces. The
correlation between the input images and the recorded
examples are calculated and used for the decision.
This meth od has the advantage of being simple but it is
significantly influenced by the variation of scale, pose and
form [32].

4) APPROACHE S / METHODS BASED ON
APPEARANCE

The objective of this approach is to determine the
significant characteristics Faces and non-faces using statistical
analysis techniques and Organized by means of distribution
models or a function Discriminating. The face or non face
classification is represented by a variable Random x (derived
from a characteristic image or vector). The metho ds used in
this context include: Neural networks, machines to vector
support SVM, hidden models of Markov HMM, Viola -Jones.
These methods use the same principle as presented in the
previous are based on models learned from a test set. These
methods have th e advantage of running very quickly but
require a long time of training. Methods in this category have
shown good Results compared to the other three types of
methods [6]. Some of the These, the method based on neural
networks of Rowley et al. [7], the Sch neiderman and Kanade
method [8] based on a Bayes naîf classifier and the famous
algorithm of Viola and Jones [9] working in real time.
III. ALGORITHM FOR THE DE TECTION OF THE FACE AND
EYES

In the first phase a face is detected. The color pixel
detection of the skin is used to extract all the entire pixels
from the skin colors of the image entered. Once they are
extracted Viola Jones is applied to detect the face. This
increases the ease of the Viola Jones technique and reduces
the time consumed.

Figure 1 : the stages of Detection of eyes by our approach

In the second phase, the detected face is extracted from the
image entry for further processing. Extract face is divided in
three sub -parts of the upper left half, the right upper half and
the lower half (Figure 2). The division of the face image is
based on the physical approximation of the location of the
eyes and the mouth on the face. The divisions are carried out
on the basis of the physical approximation of the eyes and
mouth. Figure 1 illustrates the proce ss performed in the
second phase, then the technique Viola -Jones is applied to
detect the eyes, the left eye (the upper left half) and the right
eye (t he upper right half). Figure 1 shows the organizational
chart of our approach.

1) SKIN COLOR PIXEL METHOD

For our approach, detection of the color of the skin is the
first stage. The main advantages of this technique are its
robustness, non -sensitivity to position and invariance form.
However, for this technique to work, it is essential to use a
precise color space model. Some existing color spaces are
RGB, CMY, XYZ, U VW, LSLM, L * a * b *, L, H, Y, Y, Y,
YcbCr [18, 19], the most commonly used are RGB, HIS,
YCbCr. We use the RGB color space for detection
of the color of the skin because it is the native representation
of color images and is widely used for the processin g and
storage of digital images.
The RGB color space consists of three base colors R (red),
G (green), B (blue) which can be combined to produce all
resulting colors.
Although different people have different skin colors,
studies have shown that the real di fference lies between the
intensities [20 ]. Therefore, if the luminosity is eliminated from
the representation of the color, the difference between
the colors of human skin can be reduced. In order to detect the
color of the skin, we must follow a set of r ules that have been
found to be more precise than other models.

The pixels of the RGB image are detected as skin if the
first condition is true and the rest of the conditions are used to
ensure that the components RGB must not be reconciled,
which ensures the elimination grayness.
The resulting image of this technique is a black -and-white
image which skin is converted to white and remains colors are
converted to black. The face can be detected by cutting the
largest white area conn ected with the black and white image.

2) VIOLA -JONES METHOD

Paul Viola and Michael Jones, then employed by the
Cambridge Research Laboratory of the American company
Compaq [10], publish the method that bears their name
for the first time on July 13, 2001 in International Journal of
Computer Vision (IJCV) [11]. The two authors then published
two more Articles on the methodology: a less detailed version,
presented at the Conference on Computer Vision and Pattern
Recognition (CVPR) in Decembe r 2001 [12] and Revised
version in 2004, still in IJCV [13].
The characteristics extracted by this method are inspired
by the work of Papageorgiou, Oren and Poggio, dating from
1998 [14, 11], which use features constructed from Haar
wavelets. The method i s also previous work of Paul Viola and
Kinh Tieu in another area, the Of the search for image by the
content, taking again the idea of selection of characteristics by
AdaBoost [15, 11]. Among the many methods of face
detection published at the time [16], V iola and Jones consider
in particular that of Rowley -Kanade [17]: because of its
excellent results and speed, they take it as a reference for
comparisons [11] with equivalent performance; Viola and
Jones note that detection by their method is 15 times fast er
than the Rowley -Kanade detector [11].
The method of Viola and Jones consists of scanning an
image using a n Initial size detection window 24px by 24px (in
the original algorithm) and Determine if a face is present.
When the image has been traversed entir ely, the size of the
window is increased and the scanning begins again until the
Window makes the size of the image. Increasing the size of
the window is done by multiplicative factor of 1.25. Scanning,
for its part, consists simply of shifting the window of a pixel.
This lag can be changed in order to speed up the process, but a
pixel shift ensures maximum accuracy.
This method is an approach based on appearance method ,
which consists in browsing the whole image by calculating a
certain number of characteristics in Overlapping rectangular
Figure 2 : The division of the face image . First condition

areas. It has the particularity of using Very simple but very
numerous characteristics.

3) CHARACTERISTICS

A characteristic is a synthetic and informative
representation, calculated from the values of the pixels. The
characteristics used here are the pseudohaar characteristics.
They are calculated by the difference of the pixel sums of two
or more adjacent rectangular zones.

III. EXPERI MENTS

To achieve our theoretical idea and have a practical result,
we based on open source tools to offer an application that
implements the algorithms used in our article.
The tools used are:
 A robust programming language in the field, object –
oriented op en source such as Java .
 Sophisticated and open source library to solve
mathematical complexities such as OpenCV .
 The Swing API for loading images and the realization
of different graphical interfaces.

OpenCV (for Open Computer Vision) is a free graphical
library, initially developed by Intel, specialized in the
processing of images in time real. The robotics company
Willow Garage provides support for this library
since 2008. This library is distributed under the BSD license.
NVidia announced In September 2 010 that it would develop
functions using CUDA for OpenCV [21 ].
The experiment part is as follows:
The interface in (Figure 3 ) presents the application interface,
in (Figure 4) the main menu a kind of gateway to access to
the applications of our approach from which we choose the
type of detection to use, After we get the image to be
processed (Figure 5), and then Face and eyes are detected in
(Figure 6) (Figure 7).

Figure 3: Application Interface

Figure 4 : Main Menu

Figure 5 : Image Inserted

Figure 6 : Face detected

Figure 7: Eyes detected

Another example of face detection and eye detection (Figure
9), (Figure 10 ).

Figure 8: Image Inserted

Figure 9 : Face detected

Figure 10 : Eyes detected

IV. CONCLUSION

Our approach features a new clear and important idea for
constructing a couple of image by using a single human image
based on faces detection and eye detection.
Once the couple of images are reconstructed we can apply
our 3D reconstruction met hods
The results obtained in the experimental part shows the
importance and the quality of our approach.

References

[1] Boudjellal. S. Detection and identification of personne by the biome trique
method. Master : Electronique -detection. Tizi -ouzou : Mouloud Mammeri
university , 95p,
[2] Ming -Hsuan Yang, David J. Kriegman and Narendra Ahuja. Detecting
faces images : A survey. Dans IEEE Transaction on Pattern Analysis and
Machine Intelligence, 2002, vol.24(1), 34 -58.,
[3] Commercial Security Devices [en li ne] (page consult_ee le
01/10/2013).<http ://www.commercialSecuritydevises.com/fr/d_etection -de-
visage.html>,
[4] Yang, 2002 : Yang M H, Kriegman D, Ahuja N. Detecting faces in
images: A survey. IEEE Trans Pattern Analysis and Ma chine Intelligence,
2002, 24(1) :34-58.,
[5] Cheng-Chin, Wen -Kai Tai, Mau -Tsuen Yang, Yi -Timg Huang, and chi –
Janng Huang. A novel method for detecting lips, eyes and faces in real time.
Real-Time Imaging, 9(4) : 277 -287, 2003.,
[6] Wenlong Zheng and Suchendra M. Bhandarkar. Face detection and
haking using a boosted adaptative particle _lter. Journal of visual
communication and Imog Representation, 20(1) : 9-27, 2009,
[7] H. A. Rowley, S. Baluja, and T. Kanade. Neural Network -based face
detection. IEEE Transachen on Pattern Analysis and Machine Intelligence,
20(1) : 23 -38,1998..,
[8] H. Schneidermand and T. Kanade. Probabilistic modeling of local
appearance and spaci al relation ships for object recognition. Computer Vision
and Pattern Recogniti on, IEEE Computer Society Confe rence on 0 :45, 1998..,
[9] Paul Viola and Michael Jones. Robust real -time object detection. In
Second international work shop on statistical and com putation theories of
vision, Vancouver, Canada, July 13 2001,
[10] Compaq Computer Corporation is a computer company created in 1982
by Rod Canion, Bill Mu rto and Jim Harris of Houston, Texas, USA. Compaq
is the abbreviation for COMPatibility and Quality.,
[11] Paul Viola et Michael Jones, « Robust Real -time Object Detection »,
IJCV. 2001,
[12] Paul Viola et Michael Jones, « Rapid Object Detection using a Boosted
Cascade of Simple Features », IEEE CVPR, 2001 ,
[13] Paul Viola et Michael Jones, « Robust Real -time Face Detection », IJCV,
2004,p. 137 -154.,
[14] C. Papageorgiou, M. Oren et T. Poggio, « A General Framework for
Object Detection », International Conf erence on Compu ter Vision, 1998 ,
[15] K. Tieu et P. Viola, « Boosting image retrieval », IEEE CVPR. 2000,
[16] M.-H. Yang, D. J. Kriegman and N. Ahuja, « Detecting faces in images :
A survey », IEEE Transactions on Pattern Analysis and Machine Intelligence,
vol. 24, no 1, 2002, p. 3458 ,
[17] H. Rowley, S. Baluja and T. Kanade, « Neural network -based face
detection », IEEE Transactions on Pattern Analysis and Machine Intelligence,
vol. 20,1998, p. 2238,
[18] H. X. Xinyu Wang, Xi Chen and Heng Li, "Fast and Robust Face
Detection with Skin Color Mixture Models and Asymmetric AdaBoost," Proc.
of SPIE Vol.7496 749618 -1, 2009 ,
[19] I. Hsieh, S., Fan, K., C., Lin, C." A statistic approach to the detection of
human faces in color nature scene," Pattern Recognition, 35(7), 1583 -1596
(2002) ,
[20] V. Vezhnevets, Sazonov, V. and Andreeva, A. (2003). "A Survey on
Pixel -Based S kin Color Dete ction Techniques ",
[21] http://www.tmworld.com/article/511151PUsboostdesignandsimulation :
php,
[22] Martin Willich, 2nd End -User Group Meeting on 3D Face Recognition,
February 2008, Berlin ,
[23] http ://opencvlibrary.sourceforge.net/ – OpenCV documentation and
FAQs – 25Mai 2010 – [OpenCV, 2007]A Basic Introduction to OpenCV for

Image Processing Discover Lab School of Information – Technology
Engineering – University of Ottawa January, 2007, detection
[24] B. Raytchev, H. Murase. Unsupervised face recognition by associative
chaining. Pattern Recognition 36 (2003) 245 -257
[25] J. Zhang, Y. Yan, and M. Lades. Face recognition: Eigenfaces, elastic
matching, and neural nets. Proceedings of IEEE, 85(1997) 1422 -1435
[26] D. Valentin, H. Abdi, A.J. O'toole, G.W. Cottrell. Connectionist models
of face processing: a survey. Pattern Recognition 27(9) (1994) 1209 -1230
[27] S. Pang, D. Kim, S.Y. Bang. Membership authentication in the dynamic
group by face classification using SVM ensemble. Pattern Recognition Letters
24 (2003) 215 -225
[28] J. Lu, K.N. Plataniotis, A.N. Venetsanopoulos. Face Recognition using
kernel direct discriminant analys is algorithms. IEEE Trans. on Neural
Networks 14(1) (2003) 117 -126
[29] I. Biederman. Human image understanding: Recent research and theory.
CVGIP , 32:29–73, 1985.
[30] F. Nahm, A. Perret, D. Amaral, and T. Albright. How do monkeys look at
faces? Journal of Cognitive Neuroscience , 9:611–623, 1997.
[31] M. Tistarelli. Active/space -variant object recognition. Image and Vision
Computing , 13(3):215 –226, 1995.
[32] W. Zhao, R. Chellappa, P. Phillips, and A. Rosenfeld. Face recognition:
A literature survey. ACM Computing Surveys , 35:399 – 458, 2003.