Local Region Pseudo-Zernike Moment- Based Feature Extraction for Facial Recognition of Identical Twins

Advanced Computational Intelligence: An International Journal (ACII),Vol.1, No.1, July 2014
19
LOCAL REGION PSEUDO-ZERNIKE
MOMENT- BASED FEATURE EXTRACTION
FOR FACIAL RECOGNITION OF IDENTICAL
TWINS
Zahra Ahmadi-Dastjerdi1 and Karim Faez2
1Department of Computer Engineering, Qazvin Branch, Islamic Azad University, Qazvin,
Iran
2
Department of Electrical Engineering, Amirkabir University of Technology, Tehran,
Iran
ABSTRACT
In the domain of image processing, face recognition is one of the most well-known research field. When
humans have very similar biometric properties, such as identical twins, the face recognition system is
considered as a challengeable problem. In this paper, the AdaBoost method is utilized to detect the
facial area of input image. After that the facial area is divided into some local regions. Finally, new
efficient facial-based identical twins feature extractor based on the geometric moment is applied into
local regions of face image. The used feature extractor is Pseudo-Zernike Moment (PZM) which is
employed inside the local regions of facial area of identical twins images. To evaluate the proposed
method, two datasets, Twins Days Festival and Iranian Twin Society, are collected where the datasets
includes scaled and rotated facial images of identical twins in different illuminations. The experimental
results demonstrates the ability of proposed method to recognize a pair of identical twins in
different situations such as rotation, scaling and changing illumination
KEYWORDS
Face Recognition, Local Regions, Identical Twins, Invariant Moment, Pseudo-Zernike
Moment
1. INTRODUCTION
Face is one of the best features to identify people identity from his (her) image. According
to this property, identification of facial of identical twins is critical because of the
similarity between a pair of twin. To identify identical twins, some important previous works
are listed as: in [12], the proposed face detection method contains three levels: (1) overall
appearance of the face is constructed; (2) exact geometric and structural embedment of face
with differentiating between two similar faces are performed; and (3) the third level consists
of process of skin disorders detection such as wounds. Sun et al. [16] utilized Cognitec
FaceVACS system to recognize identical twins from CASIA Multimodal Biometrics
Database. They obtained true accept rate of approximately 90% at a false accept rate greater
than 10%. In [14], the proposed fce detection method includes tree steps: (1) the proposed
method marks the face images using normal geometric methods; (2), the Euclidean distance
between a pair of markers of test image and images in dataset, are computed and compared;
and (3), the algorithm involves finding the strong similarity between the marked regions.

20
In [15], recognition of identical twins is performed using marks on the face image.
Martin et al. [3] employed DNA approach to recognize identical twins. In this paper, the
geometric moments is used to extract feature vector from input facial images of twins in order
to recognize identical twins.
This paper is organized as follows: in Section 2, feature extraction step of a face recognition
system is introduced. In Section 3, the proposed method is demonstrated. In Section 4,
experimental results are presented and the paper will be concluded in Section 5.
2. FEATURE EXTRACTION
Each face detection system contains four steps: preprocessing, face localization, feature
extraction and classification. Feature extraction is a process which is employed to collect
useful information from raw data and so this process is necessary for the pattern recognition
problems. Since the feature extraction methods are not public and depend on application,
feature extraction step may show different results. There are two different groups for the
feature extraction methods: structural features and statistical features [11][19]. In the first
group, local structure of input image takes into account where these structural features deal
with local data [7].
In the second group, the statistics-based feature extraction methods create a set of feature
vector according to the global data of input image. Extraction of irrelevant information from
facial image may create unappreciated set of feature vector elements such as hair, shoulders,
and background should be regarded in the feature extraction phase [10]. Some of statistics-
based feature extraction methods are listed as Principle Component Analysis (PCA),
Legendre Moment (LM) [13] and Zernike Moments (ZM) [20], Pseudo-Zernike Moment
(PZM) [8]. Legendre functions are Legendre differential equation. It is important that
Legendre moments are orthogonal, independent of each other and free of data redundancy.
Zernike Moment (ZM) is a set of orthogonal polynomials which are defined into a unit
disk. The ZM technique is independent of scale and rotation of face in image. The Pseudo-
Zernike Moment (PZM) is the similar to ZM but the feature vector elements extracted by
PZM is more than the feature vector extracted by ZM and so is more suitable than ZM for
recognition of identical twins. In this study, PZM is utilized to extract feature vector in order
to recognize identical twins. The PZM will be described in the next Section.
3. PROPOSED METHOD
In this paper, the goal is distinguishing of identical twins using facial image. For this purpose,
at the first step, a boosting method named AdaBoost [18] technique is employed to localize
the facial area of input image and also create subimge. In the next step, the PZM is applied on
each subimage to create feature vector elements for each subimage and this step is performed
for all images in dataset. In the final step, the feature vectors of test image is compared to
feature vectors of all images in dataset and the image of dataset with the minimum distance to
the test image is selected as the pair of test image. In the next Section, the processes of face
detection and feature extraction will be described in detail.

21
3.1. Face Detection Method
In the process of identical twins recognition, the second step is face detection. In the proposed
method, mixture of successively more complex classifiers in a cascade structure using
AdaBoost [18] is applied to detect facial area of input image using a small number of Haar-
like features [18]. After ﬁnding the facial area, an ellipse will be drowning around the main
location of face [8]. The ellipse shape includes five parameters: X0 and Y0 are the centers of the
ellipse, θ is the orientation, α and β are the minor and the major axes of the ellipse,
respectively.
Before the computation of the above mentioned parameters, geometric moments should be
described. The geometric moments of order p+q of a digital image are defined as
where p, q = 0, 1, 2, … and f(x, y) is the grey-scale value of the digital image at x and y
location. The translation invariant central moments are achieved by placing origin at the center
of the image:

22
3.2. Pseudo-Zernike Moment (PZM)
PZM is based on the geometric moment and used to create feature vector according to the
global data of an image [9]. The advantages of PZM is that its orthogonal moments are shift,
rotation, and scale invariants which are suitable for pattern recognition problems
[6][5][8][17]. Also, PZM includes several orthogonal sets of complex-valued polynomials
defined as
3.3 Creating feature vector
In this step, PZM is applied on each subimage to extract feature vector. For this purpose,
the feature vector elements are defined as:
3.4 Local Regional Pseudo-Zernike Moment (LRPZM)
The main contribution of this paper is focused on applying of PZM in a local area. For this
purpose, the obtained subimage from AdaBoost is divided into 13 subregions including an eye

23
and two eyebrow regions (for both the left and right eye); an upper, middle, lower left,
and lower right regions of the nose; and the left, middle, and right regions of the mouth.
These regions are normalized to attenuate variation in illumination. After that, the PZM is
employed in each subregion to extract a feature vector per region.
To find the pair of a twin, a set of feature vectors obtained from regions of input image will
be compared with the sets of feature vectors which are produced from the image in the
database. In the other words, the distance between the feature vector of a specific region in
input image and the feature vector of same region in an image of database (this action is
utilized for all regions). If there are 13 regions, then there are 13 distance values. The
average value of all distance values is calculated as the distance value between these two
images. Finally, the image from database with minimum distance value is considered as the
pair of input image. In the next Section, we discuss on the evaluation of the proposed method.
4. EXPERIMENTAL RESULTS
To evaluate the proposed method, two datasets is used: Twins Days Festival [2] and
Iranian Twin Society [1] with 520 and 600 pairs of identical twins images, respectively. Both
of The datasets includes facial image with different scales, rotation and different
illuminations. In Figure 1, subimages of some twin test images are shown.
The results of identical twins recognition of LRPZM is compared with PZM, ZM [20] and
LM [13]. Experiments results have been carried out in three steps according to order of
moment: (1) order n is in interval [1,6]; (2) order n is in interval [6,8]; and (3), order n is in
interval [9,10]. In this paper, N is set 10 (N=10) and j varies from 1 to 9. The error rate on the
classification of all geometric moments (LM, ZM, PZM and LRPZM) is calculated using
(15) and the results will be reported in Table 3.
Table 1. Feature vector elements based on the PZM
j value
FV j feature elements ( PZM km
) Number of
feature element
K M
4
4 0,1,2,3,4
56
5 0,1,2,3,4,5
6 0,1,2,3,4,5,6
7 0,1,2,3,4,5,6,7
8 0,1,2,3,4,5,6,7,8
9 0,1,2,3,4,5,6,7,8,9
10 0,1,2,3,4,5,6,7,8,9,10
6
6 0,1,2,3,4,5,6
45
7 0,1,2,3,4,5,6,7
8 0,1,2,3,4,5,6,7,8
9 0,1,2,3,4,5,6,7,8,9
10 0,1,2,3,4,5,6,7,8,9,10
9
9 0,1,2,3,4,5,6,7,8,9
21
10 0,1,2,3,4,5,6,7,8,9,10

24
Table 2. Feature vector elements produced by geometric moments in each experiment.
Cat. LM feature elements ZM feature elements PZM feature elements
1
n=1, m=1
n=2, m=0,2
n=3, m=1,3
n=4, m=0,2,4
n=5, m=1,3,5
n=6, m=0,2,4,6
n=1, m=1
n=2, m=0,2
n=3, m=1,3
n=4, m=0,2,4
n=5, m=1,3,5
n=6, m=0,2,4,6
n=1, m=0,1
n=2, m=0,1,2
n=3, m=0,1,2,3 n=4,
m=0,1,2,3,4 n=5,
m=0,1,2,3,4,5 n=6,
m=0,1,2,3,4,5,6
2
n=6, m=0,2,4,6
n=7, m=1,3,5,7
n=8, m=0,2,4,6,8
n=6, m=0,2,4,6
n=7, m=1,3,5,7
n=8, m=0,2,4,6,8
n=6, m=0,1,2,3,4,5,6
n=7, m=0,1,2,3,4,5,6,7
n=8, m=0,1,2,3,4,5,6,7,8
3
n=9, m=1,3,5,7,9
n=10, m=0,2,4,6,8,10
n=9, m=1,3,5,7,9
n=10, m=,0,2,4,6,8,10
n=9, m=0,1,2,3,4,5,6,7,8,9
n=10, m=,0,1,2,3,4,5,6,7,8,9,10
Table 3 shows misclassification rates of LM, ZM, PZM and LRPZM. According to the table,
higher order moments of the PZM and LRPZM extracts more information for face recognition
while low-order moments have no significant effect on the system error. Also, the table shows
high misclassification rate of LM because the LM is sensitive to the rotation of input face.
Tables 3 shows the superiority of PZM to ZM because PZM is shift, rotation and scale
invariant while the ZM is shift and scale invariant. Also the size of feature vector created by
PZM is more than the one extracted by ZM and so, the global information obtained by PZM
is sufficient for identical twins recognition. As the both of the PZM and LRPZM have the
same feature vector length, the LRPZM takes the first place in Table 3 because the LRPZM
analyse the image in small region with more detail. In Figure 2, visual results of LRPZM on
pair of identical twins of two mentioned datasets are demonstrated. In this table, the first
row shows the results of LRPZM on the Twins Days Festival dataset [2] and second row
illustrates the results of LRPZM on the Iranian Twin Society dataset [1].
Table 3. Misclassification rate of each geometric moment in different categories.
Following parameters are means as No (Number), FE (Feature Elements) and ER
(Error Rate).

25
Figure 1. Drawing ellipse around the facial area of input image.
Figure 2. Some of testing identical twins which were correctly classified by LRPZM. The first row
refers to the results on the Twins Days Festival dataset [2] and the second row is the results on the
Iranian Twin Society dataset [1].
The experimental results prove the ability of LRPZM in extraction of informative feature vector
inside the subimages of a pair of identical twins in scaling, rotation and different illumination.
5. CONCLUSIONS
In this paper, a system presented to improve the recognition of a pair of identical twins. The
proposed method is based on the Local Region Pseudo-Zernike Moment (LRPZM) as a feature
extractor to recognize a pair of identical twins. For the face detection, AdaBoost method is
applied on the facial image of twins. After that, the obtained subimage is divided into some
local regions and then the PZM is utilized to construct feature vector elements for each region.
According to the experimental results, the proposed LRPZM system is able to extract
informative feature vector from input image and also is robust to rotation and scaling and
changing illumination.
REFERENCES
[1] Iranian twin society. http://www.irtwins.com/.
[2] Twins days festival. http://www.twinsdays.org.
[3] Arias, E., MacDorman, M.F., Strobino, D.M. & Guyer, B. (2003) “Annual summary of vital
statistics2002”, Pediatrics, Vol. 112, No. 6, pp. 1215-1230.
[4] Bailey, R.R. (1993) “Automatic recognition of handwritten numerals via orthogonal moments using
statistical and neural network classifiers”, Ph.D. thesis, Dallas, TX, USA (1993), aAI9331109.
[5] Belhumeur, P., Hespanha, J. & Kriegman, D. (1997) “Eigenfaces vs. fisherfaces recognition using
class specific linear projection”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.
19, No. 7, pp. 711-720.
[6] Belkasim, S., Shridhar, M. & Ahmadi, M. (1991) “Pattern recognition with moment invariants: A
comparative study and new results”, Pattern Recognition, Vol. 24, No.12, pp. 1117-1138,
http://www.sciencedirect.com/science/article/pii/003132039190140Z.
[7] Bichsel, M. & Pentland, A. (1994) “Human face recognition and the face image sets topology”CVGIP
Image Understanding, Vol. 59, No. 2, pp. 254-261.

26
[8] Haddadnia, J., Ahmadi, M. & Faez, K. (2002) “An efficient method for recognition of human faces
using higher orders pseudo Zernike moment invariant”, In: Proceedings. Fifth IEEE International
Conference on Automatic Face and Gesture Recognition 2002, pp. 330-335.
[9] Haddadnia, J., Faez, K. & Moallem, P. (2001) “Neural network based face recognition with moment
invariants”, In Proceedings of Int. Conference on Image Processing 2001, Vol. 1, pp.1018-1021.
[10] Haddadnia, J., Ahmadi, M. & Faez, K. (2003) “An efficient feature extraction method with pseudo-
Zernike moment in RBF neural network-based human face recognition system”, EURASIP
Journal of Appl. Signal Process, pp. 890-901,
http://dx.doi.org/10.1155/S1110865703305128.
[11] Hjelms, E. & Low, B.K. (2001) “Face detection: A survey”, Computer Vision and Image
Understanding, Vol. 83, No. 3, PP. 236-274.
[12] Klare, B. & Jain, A. (2010) “On a taxonomy of facial features”, In Fourth IEEE International
Conference on Biometrics, Theory Applications and Systems (BTAS), pp. 1-8.
[13] Liao, S. & Pawlak, M. (1996) “On image analysis by moments”, IEEE Transactions on Pattern
Analysis and Machine Intelligence, Vol. 18, No. 3, pp. 254-266.
[14] Park, U., Jillela, R., Ross, A. & Jain, A. (2011) “Periocular biometrics in the visible spectrum”, IEEE
Transactions on Information Forensics and Security, Vol. 6 No. 1, pp. 96-106.
[15] Srinivas, N., Aggarwal, G., Flynn, P. & Vorder Bruegge, R. (2012) “Analysis of facial marks to
distinguish between identical twins”, IEEE Transactions on Information Forensics and Security, Vol.
7, No. 5, pp. 1536-1550.
[16] Sun, Z., Paulino, A.A., Feng, J., Chai, Z., Tan, T. & Jain, A.K. (2010) “A study of
multibiometric traits of identical twins”, http://dx.doi.org/10.1117/12.851369.
[17] Teh, C.H. & Chin, R. (1988) “On image analysis by the methods of moments”. IEEE
Transactions on Pattern Analysis and Machine Intelligence, Vol. 10, No. 4, pp. 496-513.
[18] Viola, P. & Jones, M. (2001) “Rapid object detection using a boosted cascade of simple
features”, In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern
Recognition 2001 (CVPR’01), Vol. 1, pp. I-511-I-518.
[19] Yang, M.H., Kriegman, D., Ahuja, N. (2002) “Detecting faces in images: a survey”, IEEE
Transactions on Pattern Analysis and Machine Intelligence, Vol. 24, No. 1, pp. 34-58.
[20] Zernike, v.F. (1934), “Beugungstheorie des schneidenver-fahrens und seiner verbesserten form, der
phasenkontrastmethode”, Physica, Vol. 1, No. 7, pp. 689-704.

Local Region Pseudo-Zernike Moment- Based Feature Extraction for Facial Recognition of Identical Twins

More Related Content

Similar to Local Region Pseudo-Zernike Moment- Based Feature Extraction for Facial Recognition of Identical Twins (20)

Recently uploaded (20)

Local Region Pseudo-Zernike Moment- Based Feature Extraction for Facial Recognition of Identical Twins