56 (IJCNS) International Journal of Computer and Network Security,

Vol. 2, No. 8, August 2010

The Effect of Image Compression on Face

Recognition Algorithms
Qeethara Kadhim Al-Shayea1, Muzhir Shaban Al-Ani2 and Muna Suliman Abu Teamah3
1
Al-Zaytoonah University, Department of Management Information Systems,
Amman-Jordan, 11733
kit_alshayeh@yahoo.com
2
Amman Arab University, Department of Computer Science,
Amman-Jordan, 11953
muzhir@gmail.com
3
Amman Arab University, Department of Computer Science,
Amman-Jordan, 11953
teamah@yahoo.com

Abstract: Face recognition becomes an important field via the

revolution in technology and computer vision. This paper
concentrated on recognition rate of face recognition algorithms.
The algorithms examined are: Principal Component Analysis,
Two Dimensional Principal Component Analysis in Column
Direction, Two Dimensional Principal Component Analysis in
Row Direction and Two Dimensional Two Directional Principal
Component Analysis. All these algorithms are implemented into
two environments: training environment and recognition
environment. Then a comparison between these four algorithms
with respect to recognition rate is implemented. The proposed
algorithm is implemented via Discrete Wavelet Transform
(DWT) that minimizes the images size. A complexity reduction
is achieved by optimizing the number of operations needed. This
optimization does not increase the recognition rate only, but also
reduce the execution time. A recognition rate improvement of
4% to 5% is achieved by introducing DWT through PCA
algorithms.
Keywords: Face Recognition, Recognition Rate, Image
Compression, Principal Component Analysis and Wavelet
Transform.
1. Introduction
Face recognition becomes more popular via the fast growth
in information and communication technology. It has been
introduced in many applications especially in access control
and information security that can be applied in building
access, internet access, medical records, car license plate
number, identification number, face recognition and
surveillance system and so on [1].
Face recognition system can be organized into two major
areas; the first one called face detection in which the
algorithms focus on finding the face in an image, and the
second one called face recognition, in which the algorithm
focus on recognizing the face. Recently, face recognition
becomes more important area, and many algorithms have
been developed to implement face recognition, also
researchers have spent more effort on enhancing the
performance of these algorithms in terms of quality and
recognition rate. This work aims to compare the recognition
rate between the four indicated algorithms, and also study
the effects of using DWT as compression approach via these
algorithms.
2. Literature Reviews
There are many researches related to face recognition
algorithms such as:
Wang [1] presented a structural two dimensional principal
component analysis for image recognition, which is a
subspace learning method that identifies the structural
information for discrimination.
Norouzi, Ahmadabadi and Araabi [2] presented a new
method for handling occlusion in face recognition. In this
method the faces are partitioned into blocks and a sequential
recognition structure was developed, that increase the
correct classification rate.
Sevcenco and Lu [3] presented an enhanced principal
component analysis algorithm for improving rate of face
recognition, in which modified the histogram to match a
Gaussian shaped tonal distribution in the face images.
Choudhury [4] proposed a 3D human face recognition, the
face correlation task is completed by carrying out cross
correlation between the signature functions and of the faces
and analyzing the correlation peaks, where high correlation
peak signifies true class recognition and low or no peak
signifies false class rejection.
Wang, Ding, Ding and Fang [5] developed a 2D face fitting
assisted 3D face reconstruction algorithm that aims at the
recognizing faces of different poses when each face class has
only has only one frontal training sample. This algorithm so
called Random forest Embedded active shape model, which
embeds random forest learning into the framework of active
shape model.
Sokolov et al. [6] constructed a face recognition system
using preliminary training based on sample images of object
and non objects, where the images are represented by
separate points in multidimensional space of features.
Bourlai, Kittler and Messer [7] investigated different
optimization strategies by considering both image
compression and image resolution and demonstrate that
both the system performance and speed of access can be
improved by the jointly optimized parameter setting and the
level of probe compression.
Ebrahimpour et al. [8] proposed a face models were
processed according to the human vision pathology. Three
 (IJCNS) International Journal of Computer and Network Security, 57
different sets of anthropometric points were used for
asymmetry estimations and the resulting asymmetry
measure was calculated by averaging these estimations.
Grgic, Delac and Grgic [9] described a database of static
images of human faces that proposed a tested protocol. A
simple baseline principal component analysis face
recognition algorithm was tested following the proposed
protocol, which proving their robustness and efficiency.
Zhan et al. [10] presented an evolutionary classifier fusion
method inspired by biological evolution is presented to
optimize the performance of a face recognition system.
Different illumination environments are modeled as
multiple contexts using unsupervised learning and then the
optimized classifiers are searched.
Saleh et al. [11] proposed a shift invariant pattern
recognition mechanism using a feature-sharing
hypercolumn model. In this work a share mape is
constructed to improve the recognition rate and to reduce
the memory requirements of the hypercolumn model.
Guoxing et al. [12] a local binary pattern (LPB) has been
proved to be successful for face recognition. A new LPB-
based multi-scale illumination preprocessing method was
proposed, in which this method performs better than the
existing LPB-based methods.
Sing, Thakur, Basu, Nasipuri and Kundu [13] proposed a
self adaptive radial basis function neural network based
method for high speed recognition of human faces. It has
been seen that the variations between the images of a
person, under varying pose, facial expirations, illumination,
etc. were quite high.
Kumar et al. [14] presented a PCA memetic algorithm
approach for feature selection of face recognition. PCA has
been extended by memetic algorithm, where the former was
used for feature extraction reduction and the later exploited
for feature selection.
Banerjee et al. [15] studied the performance of frequency
domain correlation filter for face recognition by considering
only the phase part of the face images, in which the
dimensions of the phase spectra were reduced by using
frequency domain principal component analysis.
Zhangand and Qiao [16] represented a new gradient Gabor
filter to extract multi-scale and multi-orientation features to
represent and classify faces. An efficient kernel fisher
analysis method was proposed to find multiple subspaces
based on both gradient Gabor magnitude and phase features,
which is a local kernel mapping method to capture the
structure information in faces.
Huang, Yi and Pu [17] proposed a new mean shifting
incremental PCA method based on the autocorrelation
matrix, which required lower computational time and
storage capacity owing to the two transformation design.
Chougdali, Jedra and Zahid [18] proposed kernel relevance
weighted discriminant analysis for face recognition which
has several interesting characteristics. Two novel kernel
functions were introduced o improve the performance of the
proposed algorithm.
3. Face Recognition Algorithms
Face recognition algorithms are implemented in many forms
to perform high efficiency of recognition. They are affected
by many factors such as recognition rate, execution time and
Vol. 2, No. 8, August 2010
recognition accuracy. In general, face recognition
algorithms have two major phases; the training stage and
the recognition stage.
3.1 Training Phase
Training phase represents the first phase where the database
of images for known people is used, and at least one image
per known person is available at the database. In this phase
the features for each known face image are extracted and
stored in the database. There are three main steps in the
training phase that shown in Figure 1, including calculating
a projection matrix from the trained images, then extracting
features from the images and finally storing these features to
be used in the recognition phase. First step concerns with
the calculation of the projection matrix, in which Face
recognition algorithm use the training images to calculate
the projection matrix. For example in Principal Component
Analysis (PCA) the projection matrix consists of
eigenvectors of the covariance matrix, where the covariance
matrix is derived from the trained images. The second step
deals with the extraction of the features from images. In
PCA each image is multiplied by the projection matrix to
formulate the eigenfaces. The eigenfaces represent the
principal component of the images, or in other words it
represents the features of the image. Finally the last step in
the training phase is to store the features that are extracted
from the trained images [1], [3].
Calculate Calculate
Store
projection feature
feature
Face matrix extraction
images as vector in
database
input of
training
phase
Projection Feature
matrix vector
Figure 1. Training phase of general face recognition
algorithm.
3.2 Recognition Phase
Recognition phase represents the second phase, in which
each new unknown face image is analyzed to obtain its
features, and then a comparison between its features and the
stored features from the training stage is performed to
identify the unknown face image. Many algorithms for face
recognition have both a training stage and a recognition
stage. This phase consists of two main steps that shown in
Figure 2, including extracting the features for the unknown
image and comparing it with the stored features. The
calculated projection matrix in the training phase is used to
extract the features of the unknown image. So in order to
extract the features from the unknown image applying PCA
algorithm, multiply the unknown elements of images by the
projection matrix which is derived from the training stage.
When the feature of the unknown image is extracted, a
comparison process is performed between the extracted
features and the stored features from the training stage.
There are many methods to do such a comparison, such as
the Euclidean distance measure. The distance between the
features of the unknown image and the stored features is
computed, and then the minimum distance corresponding to
the closest face features is selected as the matched face [17].
58 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 8, August 2010

5. Results and Discussion

Calculate Comparing
Identifying A different algorithm are implemented to present the
feature the result
Face extraction with the
the recognition rate for the PCA algorithm before and after
images as unknown
stored applying DWT. Applying DWT indicates that the
input of face images
features
recognitio recognition rate increases and also the size of the feature
n phase
Feature Decision
matrix is reduced. DWT can be implemented via different
vector making types of DWT filters such as Harr, Mallet, Daubechies and
other types.
Figure 2. Recognition stage of general face Figure 5-a shows the comparison between the recognition
recognition algorithm. rate of PCA, PCA using first level DWT and PCA using
second level DWT. The results shows that the recognition
4. Proposed System rate increase by 1.4% by using first level DWT and increase
The proposed system based on discreet wavelet transforms by 4.2% by using second level DWT.
(DWT) algorithm to minimize image size, then PCA Figure 5-b shows the comparison between the recognition
algorithm is applied for face recognition. DWT is used to rate of 2D-PCA-C, 2D-PCA-C using first level DWT and
decomposed the original image into four sub images each 2D-PCA-C using second level DWT. The results shows that
one contains some of the information of the original image. the recognition rate increase by 1.6% by using first level
The most important components of the original image are DWT and increase by 4.8% by using second level DWT.
stored in one part, which is the low low (LL) sub image that Figure 5-c shows the comparison between the recognition
represents identical features of the original image. This will rate of 2D-PCA-R, 2D-PCA-R using first level DWT and
increase the recognition rate even so it uses images of 2D-PCA-R using second level DWT. The results shows that
smaller size. Second level DWT is implemented to achieve the recognition rate increase by 1.0% by using first level
more reduction in image size, i.e. more reduction in DWT and increase by 4.2% by using second level DWT.
comparisons operations required. In addition this Figure 5-d shows the comparison between the recognition
improvement in recognition rate cases some reduction in rate of (2D)2-PCA, (2D)2-PCA using first level DWT and
image resolution as shown in Figure 3. (2D)2-PCA using second level DWT. The results shows that
Figure 3. Decomposing an image using DWT (second the recognition rate increase by 1.8% by using first level
DWT and increase by 4.6% by using second level DWT.

(a) Recognition rate of PCA before and after applying DWT

95

90 .... after applying 2nd level DWT

---- after applying 1st level DWT
percentage of the recognition rate

-.-. befor applying DWT

85

80

75

70

level).
65

For each new image after applying feature extraction

60
process, then comparing the extracted features with the 1 1.5 2 2.5 3 3.5 4 4.5 5
number of images per person
features stored from the training stage. Two possibilities will
appear: the extracted features are matched one of the stored (b) Recognition rate of 2D - PCA - C before and after applying DWT
95
images, so this leads to a matching decision, other
possibility the extracted features are not matched one of the
90 .... after applying 2nd level DWT
stored images, so this leads to a mismatching decision, in
percentage of the recognition rate

---- after applying 1st level DWT

this case the features of the new image are added as a new -.-. befor applying DWT
85
features to the database, as shown in Figure 4.
80
Trained Feature Input of
images extraction new
image
75

Yes Identify
Add the It is
new image matched matching
70
to the the feature vector
database
No
65
1 1.5 2 2.5 3 3.5 4 4.5 5
number of images per person

Figure 4. Flow chart of the face recognition algorithm.


(c) Recognition rate of 2D - PCA - R before and after applying DWT
95
90 .... after applying 2nd level DWT
percentage of the recognition rate
---- after applying 1st level DWT
-.-. befor applying DWT
85
80
75
70
65
1 1.5
(d) Recognition rate of (2D)2 - PCA before and after applying DWT
95
90 .... after applying 2nd level DWT
percentage of the recognition rate
---- after applying 1st level DWT
85
-.-. befor applying DWT
80
75
70
65
1 1.5
Figure 5. Comparison of recognition rate for different
algorithms.
6. Conclusion
This work examines the four algorithms of face recognition:
Principle Component Analysis, Two Dimensional Principle
Component Analysis in Column direction, Two
Dimensional Principle Component Analysis in Row
Direction and Two Dimensional Two Directional Principal
Analysis. An algorithm is proposed to reduce the execution
time of the face recognition algorithms and improve the face
recognition rate. The execution rate of an algorithm depends
mainly on the number of operations needed to recognize the
indicated face image.
Introducing DWT through PCA algorithms improved the
recognition rate up to 5%, also we can conclude that there
are no effective effects on applying the 2nd level DWT.
References
[1] H. Wang, "Structural Two Dimensional Principal
Component Analysis for Image Recognition", Machine
Vision and Applications, Jan. 2010.
[2] E. Norouzi, M. N. Ahmadabadi and B. N. Araabi,
"Attention Control with Reinforcement Learning for
Face Recognition under Partial Occlusion", Machine
Vision and Applications, Jan. 2010.
[3] A. Sevcenco and W. Lu, "Perfect Histogram Matching
PCA for Face Recognition", Multidimensional System
and Signal Processing, 14 Jan. 2010.
(IJCNS) International Journal of Computer and Network Security, 59
Vol. 2, No. 8, August 2010
[4] T. Choudhury, "Three Dimensional Human Face
Recognition", Optical Society of India, Vol. 38, No. 1,
pp. 16-21, 2009.
[5] L. Wang, L. Ding, X. Ding and C. Fang, "2D Face
Fitting-Assisted 3D Face Reconstruction for Pose-
Robust Face Recognition", Soft Computing, 8 Nov.
2009.
[6] M. Sokolov et al., "Face Recognition Using Lateral
Inhibition Function Features", Optical Memory and
Neural Networks, Vol. 18, No. 1, pp. 1-5, 2009.
[7] T. Bourlai, J. Kittler and K. Messer, "On Design and
Optimization of Face Verification Systems that are
Smart-Card Based", Machine Vision and Applications,
2 2.5 3 3.5 4 4.5 5 10 Feb. 2009.
number of images per person
[8] Ebrahimpour et al., "Applying Computer Stereovisio
Algorithms to Study of Correlation between Face
Asymmetry and Human Vision Pathology", Pattern
Recognition and Image Analysis, Vol. 19, No. 4, pp.
679-686, 2009.
[9] M. Grgic, K. Delac and S. Grgic, "SCface – Surveillance
Cameras Face Database", Multimedia Tools and
Applications, 30 Oct. 2009.
[10] Y. Zhan et al., "Evolutionary Fusion of Multi-Classifier
System for Efficient Face Recognition",International
Journal of Control, Automation, and Systems, Vol. 7,
No. 1, pp. 33-40, 2009.
[11] A. Saleh et al., "Feature Map Sharing Hypercolumn
Model for Shift Invariant Face Recognition", Artificial
2 2.5 3 3.5 4 4.5 5
number of images per person Life Robotics, 5-9 Feb. 2009.
[12] Guoxing et al., "An LPB-Based Multi-Scale
Illumination Processing Method for Face Recognition",
Journal of Electronics (China), Vol. 26, No. 4, Jul.
2009.
[13] J. K. Sing, S. Thakur, D. K. Basu, M. Nasipuri and M.
Kundu, "High Speed Face Recognition Using Self
Adaptive Radial Basis Function Neural Networks",
Neural Computing and Applications, 24 Feb. 2009.
[14] Kumar et al., "Feature Selection for Face Recognition:
A Memetic Algorithm Approach", Journal of Zhejiang
University Science A, Vol. 10, No. 8, pp. 1140-1152,
2009.
[15] Banerjee et al., "Illumination and Noise Tolerant Phase
Recognition Based on Eigen Phase Correlation Filter
Modified by Mexican Hat Wavelet", Journal of Optical
Society, Vol. 38, No. 3, pp. 160-168, 2009.
[16] B. Zhangand and Y. Qiao, "Face Recognition Based on
Gradient Gabor Feature and Efficient Kernel Fisher
Analysis", Journal of Neural Computing and
Applications, 4 Nov. 2009.
[17] D. Huang, Z. Yi and X. Pu, "A New Incremental PCA
Algorithm with Application to Visual Learning and
Recognition", Neural Process Letter, Vol. 30, pp. 171-
185, 2009.
[18] K. Chougdali, M Jedra and N. Zahid, "Kernel
Relevance Weighted Discriminant Analysis for Face
Recognition", Pattern Analysis Applications, 9 Apr.
2009.
60 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 8, August 2010

Authors Profile

Qeethara Kadhim Abdul Rahman Al-Shayea, has received Ph.

D. in Computer Science, Computer Science Department,
University of Technology, Iraq, 2005. She received her M.Sc.
degree in Computer Science, Computer Science Department from
University of Technology, Iraq, 2000. She has received her High
Diploma degree in information Security from Computer Science
Department, University of Technology, Iraq, 1997. She has
received B. Sc. Degree in Computer Science Department from
University of Technology, Iraq, 1992. She joined in September
(2001-2006), Computer Science Department, University of
Technology, Iraq as assistant professor. She joined in September
2006, Department of Management Information Systems Faculty of
Economics & Administrative Sciences Al-Zaytoonah University of
Jordan as assistant professor. She is interested in Artificial
intelligent, image processing, computer vision, coding theory and
information security.

Muzhir Shaban Al-Ani has received Ph. D. in Computer &

Communication Engineering Technology, ETSII, Valladolid
University, Spain, 1994. Assistant of Dean at Al-Anbar Technical
Institute (1985). Head of Electrical Department at Al-Anbar
Technical Institute, Iraq (1985-1988), Head of Computer and
Software Engineering Department at Al-Mustansyria University,
Iraq (1997-2001), Dean of Computer Science (CS) & Information
System (IS) faculty at University of Technology, Iraq (2001-2003).
He joined in 15 September 2003 Electrical and Computer
Engineering Department, College of Engineering, Applied Science
University, Amman, Jordan, as Associated Professor. He joined in
15 September 2005 Management Information System Department,
Amman Arab University, Amman, Jordan, as Associated Professor,
then he joined computer science department in 15 September 2008
at the same university.
Muna Suliman Abu-Teamah has received M.Sc in Computer
Science, Amman Arab University. (Oct 1998 to Oct 1999) Lab
Assistance in Zarqa Private University. (Oct 1999 to May 2000)
Technical & Administrator in Zarqa Private University.(Sep 2000
to Jan 2007) Teacher in Um Al-Drda'a School. (Dec 2002 to
March 2003) Teacher in Al-Dwali Center. (Dec 2008 pld assistent
in UNRWA HQA Amman. (Jan 1998 to Jan 2010) Teacher in Al
Huda-Wl-Nour Center.