Journal of Computer Science 8 (9): 1585-1593, 2012

ISSN 1549-3636
2012 Science Publications

Human Face Detection and Recognition using Web-Cam

Petcharat Pattanasethanon and Charuay Savithi
Depatment of Business Computer, Faculty of Accountancy and Management,
Mahasarakham UniversityKamreang, Kantharawichai, Mahasarakham 44150, Thailand
Abstract: Problem statement: The illuminance insensitivity that reflects the angle of human facial
aspects occurs once the distance between the object and the camera is too different such as animated
images. This has been a problem for facial recognition system for decades. Approach: For this reason,
our study represents a novel technique for facial recognition through the implementation of Successes
Mean Quantization Transform and Spare Network of Winnow with the assistance of Eigenface
computation. After having limited the frame of the input image or images from Web-Cam, the image is
cropped into an oval or eclipse shape. Then the image is transformed into greyscale color and is
normalized in order to reduce color complexities. We also focus on the special characteristics of
human facial aspects such as nostril areas and oral areas. After every essential aspectsarescrutinized,
the input image goes through the recognition system for facial identification. In some cases where
the input image from the Web-Cam does not exist in the database, the user will be notified for the
error handled. However, in cases where the image exists in the database, that image will be
computed for similarity measurement using Euclidean Distance measure from the input image.
Results and Conclusion: The result of our experiment reveals that the recognition process of 150
images in the database and 10 images from the Web-Cam provides 100% accuracy in terms of
recognition. The runtime in this case is at 0.04 sec.

Key word: Successes Mean Quantization Transform (SMQT), Neural Network (NN), Histogram
Equalization (HE), Local Binary Patterns (LBP)

INTRODUCTION greyscale and normalize the color to reduce color

complexities (Micael et al., 2007). The image or data is
One of the most challenging tasks that a facial then transformed using Successes Mean Quantization
recognition retrieval model has to tackle is the Transform (SMQT) technique and match with the
efficiency to identify the accurate match with the least database referencing the similarity measurement of
runtime result. From most video files, blurry lighting both images. Nonetheless, if the image does not exist in
has made it a disadvantage to identify facial clarity the database, the input data will be kept in the database
(Agarwal et al., 2010). In a constantly moving and the system will notify the user. On the other hand,
environment such as personal vehicles, only certain if the image exists in our database, the system will
parts of the overall facial are captured. Therefore, it is compute for similarity measurement from the
even more demanding to develop a capture model and Euclidean Distance of the image exist in our database.
recognition model from Web-Cams (Zhu et al., 2010). Crimes, homicides and assaults are literally the primary
However, in this study, we have developed a model problems of various countries. In some countries,
that provides the solution for both models to be met. social security is not provided equally to the entire
There are two critical stages for the model to follow, population due to the low financial support for such
which are the Facial Detection stage and the Facial over-prized system (Paschalakis and Bober, 2003). Our
Recognition stage. The purpose of this research is to work supports low cost and highly productive system
develop a facial recognition model which can in order to assist organizations such as government,
accurately predict or match the subject's image from retails, shopping centers, or even private businesses
our database. Input image can be either from the that require sufficient security provider. The other
camera storage or the database itself. The essential aspect that this system took advantage of is the
parts of human facial characteristics consist of eyes, difference in facial characteristics of human. The
nostril shapes, mouth shapes and skin. After the input interesting part is that we humans have different
image is uploaded, we set the limitation for its edged, biological characteristics that are hardly changed or
compute for the area of human skin, crop the facial implanted physically. Besides, it is a multitasking
shape into and eclipse shape, convert the color into system that user does not have to physically touch the
Corresponding Author: Petcharat Pattanasethanon, Depatment of Business Computer, Faculty of Accountancy and Management,
Mahasarakham UniversityKamreang, Kantharawichai, Mahasarakham 44150, Thailand
1585
 J. Computer Sci., 8 (9): 1585-1593, 2012

equipment or yield from doing other activities, since it algorithm using single image matching method. The
will detect the subject automatically that the subject model is designed to have two panels in the interface.
does not know that he or she is being captured. We can The control panel can be interpreted as two parts; the
easily agree on the benefits of facial recognition Web-Cam control i/o and the processing unit. Both
models, however, there are always some limitations in panels are processed individually and plausibly
reality that affects the flaw of the system. For instance, concurrent. The first part receives an input image
interrupt signals from images, low resolution images through the camera, which is further described in Fig.
and facial reflex or expressions can cause inaccuracy in 1a as a Startup GUI. The Start Cam button indicates
the retrieving process. These scenarios are being Web-Cam activation. The square frame is design to
handled through several research approaches. In our surround the facial area as to relocate the prospective
research, the system recognizes only the images with area and separate the facial area and the background.
frontal facial expression that are stored in web cameras The square frame creates a 10-20 sec. delay for the
and our standard database. image in order to capture the facial expression as
Related works: Slight changes to the sensor and preferred. After the Cap CAM button is pressed, the
illumination insensitivity in the image are some interrupt signal is sent to our Web-Cam, which has
examples of the many interactions that technically now stopped its task since the prospective image is
affect the entire accuracy of the pattern recognition. obtained. As shown in Fig. 1b, this process captures
Since pattern recognition can be engaged and only the facial part of the image and transforms its size
implement to several obligation such as criminal facial to reduce the storing area. The size of a normalized
matching and individual identification, a wide range of image in our research is initially set to 100140 pixels,
research has been conducted assist the above as shown in Fig. 2.
mentioned issues. For instance, some research The square frame that crops the entire face
implemented Histogram Equalization (HE), Local begins once the position of the face is set to (Sx, Sy)
Binary Patterns (LBP) (Lahdenoja et al., 2005) and andthe both eyes are set to (X1,Y) and (X2,Y), as
Modified Census Transform (MCT) (Froba and Ernst, demonstrated in Fig. 2a. The width (Wi) and Height
2004) to develop a pattern recognition model. HE is a (Hi) of the image can be computed from Eq. 1 and 2,
computationally expensive operation in comparison to respectively:
LBP and MCT, however, LBP and MCT are typically
restricted to only extract binary patterns in a local area. Wi = 2x + | x1 = x2 | (1)
The Successive Mean Quantization Transform (SMQT)
(Micael et al., 2007) can be viewed as a tunable Hi = 2y (2)
tradeoff between the number of quantization levels in
the result and the computational load. In this study the The GUI processing unit works accordingly with
SMQT is used to extract features from the local area of the flowchart describe in Fig. 3.
an image. Derivations of the sensor and illumination The second part is the identification window,
insensitive properties of the local SMQT features are which recognizes the output image by going through
presented. Pattern recognition in the context of the database with the eclipse shape frame. Later,
appearance based face detection can be approached in SMQT and Eigen face techniques are applied to the
several ways (Yang et al., 2002). Techniques proposed image. In order to compare the similarity measurement
for this task are for example the Neural Network (NN) with the database image, our reference point is based
(Rowley et al., 1998), probabilistic modelling on the value of Euclidean Distance. This will also
(Schneiderman and Kanade, 1998), cascade of boosted
thoroughly check if the person from the input image
features (AdaBoost) (Viola and Jones, 2001), Sparse
Network of Winnows (SNoW) (Gundimada and Asari, exists in database or not. Otherwise, the user is asked
2005), combination of AdaBoost and SNoW (Osuna et to directly add this image into the database, including
al., 1997) and the Support Vector Machine (SVM) the date and location which that image is found. The
(Ruan andYin, 2009). facial identification GUI window and flowchart
diagram of the system are demonstrated as in Fig. 4
MATERIALS AND METHODS and 5, respectively.

Face detection and recognition system: No matter if Face detection and face recognition technique: In
the input is a single image or a video file, the algorithm this research, we propose a facial detection and facial
that pattern recognition uses is almost similar. In video recognition technique which increases the accuracy and
files, matching images are processed by digitizing the preciseness of the results. The detection and recognition
image frame by frame. Since pattern recognition in process are based on two renowned techniques called
video files can be implemented using single image Successive Mean Quantization Transform (SMQT) and
matching, in this research, we provide the matching Spare Network of Winnows (SNOW).
1586
 J. Computer Sci., 8 (9): 1585-1593, 2012

(a)
Fig. 3: GUI Input Image Flowchart explains briefly
how our GUI processes the input image

Fig. 4: GUI of Identify FaceFrom Data Base

(b)

Fig. 1: Capturing Window (a)Startup GUI (b)

Captured Image

Fig. 5: Flowchart of the GUI of Identify FaceProcess

(a) (b)
The two techniques contribute to a pattern, which store
Fig. 2: Facial capturing and saving to database(a) details of the facial part through data transformation
Dimension on face (b) Limitation face sizing and separates it from the background. The second part
1587
 J. Computer Sci., 8 (9): 1585-1593, 2012

uses Eigenface to compute for the coefficient within Can be achieved using the non-face table hxw ,
nonface
special characteristics from the facial part, which
different people may possess different facial structure face
the face table h xw and defining a threshold for . Since
or essential characteristics from others.
both tables work on the same domain, this implies that
Local SMQT feature and split up SNOW classifier: one single lookup-table:
SMQT and SNOW are initially proposed by (Micael et
al., 2007) for the purpose of facial detection in a high h x = h nonface
x h face
x (7)
speed manner without being affected by the
surrounding error signals such as illumination insensitivity Can be created for single lookup-table
or sensor variation. SMQT extracts pixels in local area of classification.
the image. The steps are described as below. Let the training database contain i = 1, 2,..., N
Let x be one pixel and D(x) be set of |D(x)| = D feature patches with the SMQT features Mi(x) and the
pixels from a local area in an image. Consider the corresponding classes ci (face or nonface). The nonface
SMQT transformation of the local area Eq. 3: table and the face table can then be trained with the
Winnow Update Rule (Gundimada and Asari, 2005).
SMQTL : D ( x ) M ( x ) (3)
Initially both tables contain zeros. If an index in the
table is addressed for the first time during training, the
where, x is a pixel D(x) is a set of x in a local area in
value (weight) on that index is set to one. There are three
the image M(x) is yields a new set of values.
training parameters; the threshold , the promotion
Once we consider the brightness in image I(x),
which consists of Reflectance: R(x), illuminance: E(x). parameter >1 and the demotion parameter 0<<1.
Additionally, the influence of the camera can be x ( M ( x ) ) andCiis a face then promotion
If h face
x w
modeled as a gain factor gand abias term b. Thus, a
model of the image can be described by Eq. 4: is conducted as follows Eq. 8:

I ( x ) = g.E ( x ) .R ( x ) + b x (M(x)) = a h x (M i (x)), x w.

h face face
(4) (8)

Where: where, a is the promotion parameter.

If Ciis a non-face and h face x ( M ( x ) ) > then
g = A gain factor
b = A bias of brightness of the image xw

demotion takes place Eq. 9:

In order to design a robust classifier for object
x ( M ( x ) ) = .h x ( M ( x ) ) , x w.
detection the reflectance should be extracted since it h face face
(9)
contains the object structure. In general, the separation
of reflectance and illuminance is ill pose problem. A
where, is the demotion parameter
common approach to solving this problem involves
This procedure is repeated until no changes occur.
assuming that E(x) is spatially smooth. Further, if the Training of the non-face table is performed in the same
illuminance can be considered to be constant in the manner and finally the single table is createdaccording
chosen local area the E(x) is given by: to Eq. 7.
E ( x ) = E, x D (5) Eigen face process: Eigen face Technique was
originally proposed by (Kirby and Sirowich,1990). Any
Given the validity of Eq. 5, the SMQT on the local the facial vector X can be compressed and reconstructed
area will yield illumination and camera-insensitive from the total sum of Mean Image Vector (Xm) and a set
features this implies that all local patterns which of image u0u1uL-1. The relation of this equation is
contain the same structure will yield the same SMQT described as shown in Eq. 10:
features for a specified level L. The number of possible
patterns using local SMQT features will be (2 L)D. L 1

Split up SNOW classifier: SNOW (Spare Network of x = x m + a i U i (10)

i=0
Winnows) has a special task for recognizing pixels
in local area. It can also create a table to extract Later, (Zhujie and Yu, 1994) applied this method
different patterns, as elaborated below. Consider a with facial recognition assessments by evaluating the
patch W of the SMQT features M(x), then a coefficient in Eq. 10. The algorithm to compute for
classifier Eq. 6: Eigen face is described below.
Determine that set S has a member equivalent to
= h
xw
nonface
x (M(x)) h
xw
face
x (M(x)) (6) M images. Each image transforms into an image vector
at size N:
1588
 J. Computer Sci., 8 (9): 1585-1593, 2012

S = {X1 , X 2 , , X m } (11) Once we have found the eigenvector, Vl, Ul Eq. 19:
M
Find the average of the members' facial images () u l = v lk k l = 1,..., N (19)
k =1
Eq. 12:
Matching procedure: A new face image (X) is
1 M
= Xn
M n =1
(12) transformed into its Eigen face components. First we
compare out input image without mean image and
multiply their difference with each eigenvector of the
where, M is the members of facial images. X matrix. Each value would represent a weight ( k)
Then find the subtraction or difference ( i) ofthe and would be saved on a vector () Eq. 20 and 21:
input image (Xi) and Mean Image ( i) as shown in
Eq. 11 and 13: k = u Tk ( X ) (20)

i = Xi i (13) T = 1 , 2 ,..., m (21)

We seek a set of M orthonormal vector, Un, which We now determine which face class provides the
best describes the distribution of the data(n). The kth best description for the input image. This is done by
vector, Uk, is chosen such that Eq. 14: minimizing the Euclidean distance(k) Eq. 22:
1 M
( U Tk n ) k =|| k ||2
2
k = (14) (22)
M
The input face is consideredto belong to a class if
Is a maximum, subject to Eq. 15: k is bellowingan established threshold . Then the
face image is considered to be a known face. If the
1 if l = k difference is above the given threshold, but bellow a
U Ti U k = ik = (15) second threshold, the image can be determined as a
0 otherwise
unknown face. If the input image is above these two
where, ik is a maximum facial or otherwise thresholds, the image is determined NOT to be a face.
Below are some parts of the command in m-file for
Eigen face Algorithm.
Note: Uk and k are the eigenvector and eigenvalues
of the covariancematrix C. Function[m, A, Eigenfaces]=EigenfaceCore(T)
We obtain the covariance matrix Cin the following % Calculating the mean image
manner Eq. 16: m = mean(T,2); % Cal. the averface image
Train_Number = size(T,2);% Cal.the dev. of each im.
1 M from mean im.
C= nTn = A.A T
M n =1
(16)
A = [];
fori = 1 : Train_Number
where, A = {1 2 ,..., n } AT is Transpose of A. temp = double(T(:,i)) -m; % Cal. the diff. im.
A = [A temp]; % Merging all centered images
We can solve for the dimensional eigenvectors in end
this case by taking appropriate linear combinations of L = A'*A; % L is the surrogate of COV matrix
the face images i. C=A*A'.
Consider the eigenvectors vi of ATA such that Eq. 17: [V D] = eig(L); % Diagonal elem.of D are the eigen.
% Sorting and eliminate
A A T vi = i vi (17) L_eig_vec = [];
fori = 1 : size(V,2)
Premultiplying both sides by A, we have Eq. 18: L_eig_vec = [L_eig_vec V(:,i)];
end% Cal.the eigenvectors of covariance matrix 'C'
A A T A vi = i A vi (18) Eigenfaces = A * L_eig_vec; % A: centered im.
Vectors
From which we see that A vi are the eigenvectors
of C = AAT. Following these analysis, we construct RESULTS
the MM matrix L= ATA, where Lnm = mTn and
find the M eigenvectors, vi, of L. These vectors There are 150 standard frontal face images in all
determine linear combinations of theM training set face for the facial detection and recognition process. Some
images to form the eigenfaces Ui. standard frontal image set shown in Fig. 6.
1589
 J. Computer Sci., 8 (9): 1585-1593, 2012

Fig. 6: Examples of images stored in the database

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

Fig. 7: The resultsof facial detection and recognition with frontal faced and skin coloration (a) Frontal facial
(b)Matching result (c) Analysis area (d) Frontal facial (e)Matching result (f)Analysis area(g) Frontal facial
(h) Matching result (i)Analysis area
1590
 J. Computer Sci., 8 (9): 1585-1593, 2012

(a) (b) (c)

Fig. 8: The resultsof facial detection and recognition with non-frontal faced images (a) Side face (b) Matching result (c)
Analysis area

(a)

(b)

(c)

(d)

Fig. 9: The experimental results of the facial recognition system with three facial gestures (a) GUI 1 (b)Query face (c)
Matching (d) Analysis area
1591
 J. Computer Sci., 8 (9): 1585-1593, 2012

Table 1: Summary of precision in face recognition performance system DISCUSSION

No. of images Precision (%) Processing time (sec)
30 100 0.011 This research is the original model for facial
50 100 0.024 recognition through Web cam processing. Although the
100 100 0.024 result comes with low-resolution image, the accuracy of
150 100 0.034
160 100 0.040
the recognition model is at its best for retrieval and
recognition algorithm. Several implementations can be
adapted through this type of application such as in cases
Images from the database are settled for searching ofsecurity purposes or facial identification purposes.
algorithm and retrieval, while new images can be
obtained from the Web-Cam, which has a total of 10 CONCLUSION
images. In all, our database stores a total of 160 facial
This study represents a facial detection and
images in a standard frontal manner.The evaluation of
recognition model with two windows working in
our model is based on two tests. The first test uses GUI parallel and independently. The first window detects
Window 2 to select images from the database and the and stores facial images through Web-Cams. The
images from certain references such as frontal faced second window does the matching and identification of
image, frontal faced image with different skin the facial obtained from window 1. However, if the
coloration and non-frontal faced images. Figure 7 and 8 facial detected in window 1 does not exist in the
manifest the out of GUI Window 2 procedure. The database the input image will be automatically stored
second part of this detection and recognition model is directly in our database as a reference for the next input
tested by detecting images from the camera before image. The recognition process implements SMQT and
facial identification. The first step, we compared three SNOW techniques together to work concurrently with
eigenface computation. We tested this system with 150
types of image; frontal faced and non-frontal faced
frontal faced images together with facial images stored
image with black background, frontal faced images in the Web-Cam. The system reveals 100% recognition
with illuminance background and non-frontal faced accuracy. The resultsobtained from this experiment
images. The results of GUI windows 1 and GUI arebelieved to be another alternative path for facial
windows two are demonstrated as in Fig. 9. recognition in the future.
ACKNOWLEDGEMENT
System accuracy evaluation: We have tested our
system by classifying the facial images into 5 This research is financially supported by the
subcategories without repetition of same images Faculty of Accountancy and Management,
occur. The subcategories are divided into 30, 50, Mahasarakham University, Mahasarakham, Thailand.
100, 150 and 160, respectively. In the testing
REFERENCES
session, 10 images are chosen arbitrarily to perform
a facial recognition process. We can compute the Agarwal, M., N. Jain, M. Kumar and H. Agrawal,
Similarity (S) based on the Precision (P) value 2010. Face recognition using principle component
shown in Eq. 23 and 24 respectively. The results of analysis, eigenface and neural network.
the system are shown in Table 1: Proceedings of the International Conference on
Signal Acquisition and Processing, Feb. 9-10,
J t,k i,k
J IEEE Xplore Press, Bangalore, pp: 310-314. DOI:
= s(J ,J )=
Z Z
s i,k t,k i,k z
(23) 10.1109/ICSAP.2010.51
(J ) (J )
2 2
t,k i,k Froba, B. and A. Ernst, 2004. Face detection with the
Z Z
Z Z modified census transform. Proceedings of the 6th
IEEE International Conference on Automatic Face
Number of relavant image retrived and Gesture Recognition, May 17-19, IEEE
Precision = 100% (24) Xplore Press, pp: 91-96. DOI:
The total number of imag eretrieval
10.1109/AFGR.2004.1301514
Gundimada, S. and V. Asari, 2005. An improved snow
From Table 1, the first column is the amount of based classification technique for head-pose
images in our experiment. The second column denotes estimation and face detection. Proceedings of the
the accuracy for facial recognition. The first row 34th Conference on Applied Imagery and Pattern
demonstrates the testing process based on 30 facial Recognition Workshop, Dec. 1-1, IEEE Xplore
images out of the total of 160 images in all. The result Press, Washington, DC., pp: 99-99. DOI:
indicates 100% recognition. 10.1109/AIPR.2005.16
1592
 J. Computer Sci., 8 (9): 1585-1593, 2012

Kirby, M. and L. Sirovich, 1990. Application of the Ruan, J. And J. Yin, 2009. Face detection based on
karhunen-loeve procedure for the characterization facial features and linear support vector machines.
of human faces. IEEE Trans. Patt. Anal. Mach. Proceedings of the International Conference on
Intell., 12: 103-108. DOI: 10.1109/34.41390 Communication Software and Networks, Feb. 27-
Lahdenoja, O., M. Laiho and A. Paasio, 2005. 28, IEEE Xplore Press, Macau, pp: 371-375. DOI:
Reducing the feature vector length in local binary 10.1109/ICCSN.2009.76
pattern based face recognition. Proceedings of the Schneiderman, H. and T. Kanade, 1998. Probabilistic
IEEE International Conference on Image modeling of local appearance and spatial
Processing, Sept. 11-14, IEEE Xplore Press, pp: relationships for object recognition. Proceedings of
914-917. DOI: 10.1109/ICIP.2005.1530205 the IEEE Computer Society Conference on
Micael, N., J. Nordberg and I. Claesson, 2007. Face Computer Vision and Pattern Recognition, Jun.
detection using local SMQT features and split up 23-25, IEEE Xplore Press, Santa Barbara, CA., pp:
snow classifier. Proceedings of the IEEE 45-51. DOI: 10.1109/CVPR.1998.698586
International Conference on Acoustics, Speech and Viola, P. and M. Jones, 2001. Rapid object detection
Signal Processing, Apr. 15-20, IEEE Xplore Press, using a boosted cascade of simple features.
Honolulu, HI, pp: II-589-II-592. DOI: Proceedings of the 2001 IEEE Computer Society
10.1109/ICASSP.2007.366304 Conference on Computer Vision and Pattern
Osuna, E., R. Freund and F. Girosi, 1997. Training Recognition, Dec. 08-14, IEEE Computer Society,
support vector machines: An application to face pp: 511-518. DOI: 10.1109/CVPR.2001.990517
detection. Proceedings of the IEEE Computer Yang, M.H., D. Kriegman and N. Ahuja, 2002.
Society Conference on Computer Vision and Detecting faces in images: A survey. IEEE Trans.
Pattern Recognition, Jun. 17-19, IEEE Xplore Patt. Anal. Mach. Intell., 24: 34-58. DOI:
Press, San Juan, pp: 193-199. DOI: 10.1109/34.982883
10.1109/CVPR.1997.609310 Zhu, J., Y. Lao and Y.F. Zheng, 2010. Object tracking
Paschalakis, S. and M. Bober, 2003.A low cost FPGA in structured environments for video surveillance
system for high speed face detection and tracking. applications. IEEE Trans. Cir. Syst. Video
Proceedings of the IEEE International Conference Technol., 20: 223-235. DOI:
on Field-Programmable Technology, Dec. 15-17, 10.1109/TCSVT.2009.2031395
Tokyo, Japan, pp: 214-221.DOI: Zhujie and Y.L. Yu, 1994. Face recognition with
10.1109/FPT.2003.1275750 eigenfaces. Proceedings of the IEEE International
Rowley, H.A., S. Baluja and T. Kanade, 1998. Neural Conference on Industrial Technology, Dec. 5-9,
network-based face detection.IEEE Trans. Patt. IEEE Xplore Press, Guangzhou, pp: 434-438.
Anal. Mach. Intell., 20: 203-208. DOI: DOI: 10.1109/ICIT.1994.467155
10.1109/34.655647

1593