141
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
THE HUMAN EAR RECOGNITION BASED ON PHASE-
BASED MATCHING ALGORITHM
Muthukumar Arunachalam
Department of Electronics and Communication Engineering
Kalasalingam Academy of Research and Education.
Krishnankoil, (India).
E-mail: muthuece.eng@gmail.com ORCID: https://orcid.org/0000-0001-8070-3475
Recepción:
05/12/2019
Aceptación:
19/12/2019
Publicación:
23/03/2020
Citación sugerida:
Arunachalam, M. (2020). The human ear recognition based on phase-based matching algorithm.
3C Tecnología. Glosas de innovación aplicadas a la pyme. Edición Especial, Marzo 2020, 141-157. http://doi.
org/10.17993/3ctecno.2020.specialissue4.141-157
Suggested citation:
Arunachalam, M. (2020). The human ear recognition based on phase-based matching algorithm.
3C Tecnología. Glosas de innovación aplicadas a la pyme. Edición Especial, Marzo 2020, 141-157. http://doi.
org/10.17993/3ctecno.2020.specialissue4.141-157
142 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
ABSTRACT
In this modern era, biometric play a vital role in authentication process for an individual
identication. Broadly biometrics has been classied into anatomical and behavioral
characteristics. Among the many biometrics, this paper focus on new emerging biometric
named as Ear biometrics. In this paper, Phase-only Correlation (POC) and Band-limited
Phase Only Correlation (BLPOC) has been proposed for ear recognition. The phase
information has been obtained by calculating similarity between input and query ear image
using 2D-Discrete Fourier Transform (DFT) and auto correlation function. Finally the
phase-based on image matching have being success implemented for human ear recognition
endeavor. The experimental resultant eect of proposed algorithm has been performed
using IIT Delhi ear database.
KEYWORDS
DFT, Phase Only Correlation (POC), Band-Limited Phase Only Correlation (BLPOC).
143 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
1. INTRODUCTION
In impressive of last few years, Human Ear recognition has been becoming a very attractive
in biometric authentication. The important reasons behind human ear biometric over other
biometric modalities are smaller in size, very stable shape has proven by clinical observation
(Rutty, Abbas, & Crossling, 2005), uniform colors, and does not aected by any expression
like face or changes in age or position or rotation (Malathy, Annapurani, & Sadiq, 2013). The
ear is a passive biometric as it can easily be captured from far distance, even if the person
is not fully cooperative. Hence ear recognition is much interest for the researchers dealing
with numerous applications in commercial and governmental, such as law enforcement,
security systems and forensics (Jain, Ross, & Pankanti, 2006). Therefore, ear biometric is a
deserved preference for it providing a ne avail between cost and accuracy.
Fossa
Helix
Antitragus
Lower
Antihelix
Antihelical fold
Crus
antihelixis
Lobule
Incisura
intertragica
Tragus
Crus Helixis
Upper
& lower
concha
Figure 1. Ear Image.
The Figure 1 shows the ear image of an individual. It is stored in the IIT Delhi database
and can be used for the task of identication through its unique characteristics. The human
outer ears parts are formed by various terminologies include Helix, the Lobe, Antihelix, the
tragus, concha along with an antitragus.
Recently, several matching algorithms have been adapted for the ear recognition process.
Anyhow, these algorithms possess various limitations which are explained in this section.
Primarily, the holistic matching method utilized from global features to extract the complete
ear. Bustard et al. arrange the ear dataset by considering the planar surface. It was registered
as homography transform designed of Scale Invariant Feature Transform from trails
144 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
(Bustard & Nixon, 2010). This method was strong to conditions in the appearance of pose
variation and occlusion.
Ultimately, the matching algorithms with local feature have been adopted for identication.
So Nanni et al. determined to decrease the dimension by usual sub-window of feature
extraction with a Gabor Filter along with LaplacianEigenmaps (Zhu, Jia, & Liu, 2009).
They also used the Sequential Forward Floating Selection to sort out the nest matching
feature with concord trained. However the function of Laplacian Eigen maps could drop
datas extremely. Yazdanpanah et al. designed a region of covariance matrix through well-
organized plus strong ear caption, which is strong to clarication with fake dierence
(Miyazawa, Ito, Aoki, Kobayashi, & Katsumata, 2006). In the following year, Yuan et al.
ear region has extracted to segment by conserving from neighborhood impact and the
particular region has chosen by the recognition rate (Zhang, Zhang, & Zhang, 2009). This
method could keep away from the occlusion problem other than shortcoming from multi-
fake problem.
2004-2006
• Finger print
• Iris
2008
• Finger vein
• Retinal
• Palm print
• Finger knuckle print
2011
• Finger knuckle print
2012
• Finger knuckle print
2013
• Finger knuckle print
2015
• Finger print
2004-2006
• Finger print
• Iris
2008
• Finger vein
• Retinal
• Palm print
• Finger knuckle print
2011
• Finger knuckle print
2012
• Finger knuckle print
2013
• Finger knuckle print
2015
• Finger print
Figure 2. Different biometrics images used by POC and BLPOC.
The POC and BLPOC matching algorithm play the major role in image processing or
pattern matching. There are many benets in using Phase based matching algorithm such
as simple to construct score, easy to implement, nest for multiple test and close to real-
time performance. The Figure 2 shows dierent biometrics images are used by POC and
BLPOC algorithm over the years for instance nger vein, nger print, iris, retinal, palm
print as well as nger knuckle print except ear biometric. Therefore, this paper proposes
human ear recognition based on phase-base matching algorithm.
145 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Gallery Images
2D-DFT and Cross
Correlation Spectra
Phase
Component
Phase
Component
IIT Delhi Ear
Database
Enrollment process
Verication process
Phase
Matching
Matching Score
2D-DFT and Cross
Correlation Spectra
Query Image
Figure 3. Proposed system.
The Figure 3 shows the proposed system of this paper. This proposed system involves two
processes namely (i) Enrollment process and (ii) Verication process. First, enrollment process
carries out the process that involves feature extraction of POC and BLPOC technique.
This feature extraction minimizes the ear images into mathematical phase information.
This information will be stored in database to establish the authentication of ear images.
At the same time, verication process also arrive the same process for extracting features
of POC and BLPOC technique which generates the phase information. This feature
extraction converts the ear images into mathematical phase information. These both phase
information’s are aorded by decision making subsidiary. The decision making subsidiary
used to provide a nal conclusion (i.e. genuine or imposter output). The output of decision
making is a potential matching score by matching identities.
The remainders in this paper are structured as follows: The second section discusses about
survey of the literature which gives brief discussion about existed POC and BLPOC. The
third section outlines the proposed work and its implementation of Human Ear Recognition.
The fourth section discusses about presents the experimental results of proposed system.
The conclusion has been described in the section 5.
2. RELATED WORK
In this related work, discusses the detailed functions of phase-base on image matching
algorithms (i.e. POC and BLPOC). The POC has been designed to nd the correlation
146 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
within two images (i.e. dierent images or similar images). The purpose of POC is to
frame-up only phase feature and to isolate the magnitude feature of images for accuracy
in image matching. During the trial time of POC with two similar images provides distinct
sharp correlation peak and the correlation peak drops when two images are dierent
(Takita, Muquit, Aoki, & Higuchi, 2004). The experimental results of phase correlation
demonstrate high robustness and accuracy in the pattern matching and the image
registration (Ito, Nakajima, Kobayashi, Aoki, & Higuchi, 2004). This has been achieved
with direct estimation in Fourier domain through 2D phase dierence data set (Miyazawa
et al., 2006). This kind of POC correlation is more convenient and more protable than
other conventional correlation. The most remarkable advantages of POC are it provides
better discrimination for similar images and high accuracy translational of displacement
among the illustrations (Nanni & Lumini, 2007). But for most part of phase function
algorithm cannot carry the nonlinear deformation of illustrations (Rutty et al., 2005). The
BLPOC is very crucial method for correlation in optimum band limit for phase based
recognition. In POC function, every frequency components are concerned. Though, high
frequency components tend to highlight and it probably prone to noise. Hence, worthless
high frequency components are eliminated by setting a band-limited while calculating the
cross-power spectrum (Yazdanpanah & Faez, 2010). Also, in order to carry the nonlinear
deformation, it employs BLPOC matching algorithm. The BLPOC be sucient to
estimate accurate overall correlation (Yuan, Wang, & Mu, 2010). The modied version of
POC is BLPOC which is committed to evaluate match between images due to its better
performance than the POC function (Zhang, Zhang, Zhang, & Guo, 2012). The proposed
BLPOC perform (Zhang, Zhang, Zhang, & Zhu, 2011) committed to biometric recognition
tasks. Several authors attempted dierent modalities of biometric recognition by using
POC and BLPOC are outlined in the given Table 1.
Table 1. Outline about Phase-Based Function matching algorithm in various biometrics.
Author Feature Trait (Images)
(Takita et al., 2004). POC and BLPOC Finger Print
(Ito et al., 2004). POC and BLPOC Iris
(Miyazawa, Ito, Aoki,
Kobayashi, & Nakajima,
2005).
POC and BLPOC Iris
(Zhu et al., 2009). POC and BLPOC Palm print
147 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Author Feature Trait (Images)
(Mahri, Suandi, & Rosdi,
2010).
POC and BLPOC Finger Vein
(Miyazawa et al., 2006). POC and BLPOC Retinal
(Yuan et al., 2010). POC and BLPOC FKP
(Zhang et al., 2011). POC and BLPOC FKP
(Zhang et al., 2012). POC and BLPOC FKP
(Aoyama, Ito, & Aoki,
2014).
POC and BLPOC FKP
Human Ear Recognition Using Phase-Based Function Matching.
3. METHOD
In this section, POC and BLPOC are discussed and derived in detail.
Processing Stage
Matching Stage
Geniune output Imposter output
Gallery Image
g(x, y)
Consider Gallery image
has been shifted by some
position from Query image
g(x, y) = f(x - x
0
, y - y
0
)
Query Image
f(x, y)
POC
r
xy
(b
1
, b
2
)
BLPOC
r
xy
a
1
a
2
(b
1
, b
2
)
2D-DFT and Cross
Correlation Spectra
Inverse 2D-DFT
Threshold
of BLPOC
Figure 4. Overall structural outline of the proposed design of this paper.
The Figure 4 shows the overall structural outline of proposed design of this paper. This
method is discussed in next section 3.1 and 3.2 in detail.
3.1. PHASE-ONLY FUNCTION/CORRELATION (POC)
This section discussed about the principal of POC. The POC is some time called as “Phase-
Only Function”) (Zhang et al., 2009). Assume 2 M x N images, g x
1
, x
2
and the f(x
1
, x
2
),
where consider that the basis ranges are
and in
support of mathematical ease, and therefore M = 2M+1 and N = 2N+1. The conversation
can be simply general towards non-nullifying basis ranges by rule of 2 image sizes. Let F(a
1
,
148 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
a
2
)() and G(a
1
, a
2
) denote the 2D-DFTs of g(x
1
, x
2
) and, correspondingly. According to the
description of DFT (Zhang et al., 2009), F(a
1
, a
2
) and G(a
1
, a
2
) be known through
(1)
(2)
Consider
Consider image is shifted by x
0
and y
0
portion, then x
0
= y
0
= 0 the both is same. Now
applying spatial shifting property,
(3)
(4)
For nding Correlation of, using equation
(5)
Correlation phase spectra
(6)
Sub 6 and 7, so we get,
(7)
2DIDFT—
(8)
Take 2D-IDFT for equation 7,
149 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
(9)
Put equation 8 in 10,
(10)
(11)
(12)
Respectively, where a
1
=-M
1
....M
1
, a
2
=-M
2
....M
2
, , and
denotes . A
F
(a
1
,a
2
) and A
G
(a
1
,a
2
) are amplitude, and and
are phase. The normalized cross power spectrum is set by
(13)
Where G(a
1
,a
2
) is the complex conjugate of G(a
1
,a
2
)and θ(a
1
,a
2
) denotes the phase dierence
θ
F
(a
1
,a
2
)-θ
G
(a
1
,a
2
). The POC function is the 2D IDFT of and is
set by
(14)
Where
denotes . When 2 images are alike, their POC perform provides
a denite quick peak. When 2 images aren’t alike, the height drop noticeably. The height of
peak provides a decent likeness for match the image, with therefore the position of height
shows change of location displacement among the illustration (Zhang et al., 2009).
150 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
(a) (b) (c) (d) (e) (f)
Figure 5. (a) Gallery image, (b) Query image, (c) POC for the same ear images, (d) Gallery image, (e) Query
image and (f) POC for different ear images.
3.2. BAND-LIMITED PHASE ONLY FUNCTION/CORRELATION (BLPOC)
Consider to facilitate the range of the intrinsic frequency band be set through a
1
= -a
1
...
a
1
and a
2
= -a
2
... a
2
, where 0 ≤ a
1
≤ M
1
and 0 ≤ a
2
≤ M
2
. Therefore, the valuable size of
frequency spectrum be set through L
1
= 2 a
1
+ 1 and L
2
= 2 a
2
+ 1 (Miyazawa et al., 2006).
The BLPOC is known by means of
(15)
Where x
1
= -a
1
... a
1
, x
2
= -a
2
... a
2
and denote . Remind that the highest value
of correlation peak of BLPOC is constantly normalized to 1 and it not lean on L1 and L2
(Nanni & Lumini, 2007).
(a) (b) (c) (d) (e) (f)
Figure 6. (a) Gallery image, (b) Query image, (c) BLPOC for the same ear images, (d) Gallery image, (e) Query
image and (f) BLPOC for different ear images.
151 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
4. EXPERIMENT AND DISCUSSION
The intended method be extensively investigated using an Ear images database provided
by IIT Delhi. It is contained of the ear images, where collected from the students and
sta at IIT Delhi (Zhu et al., 2009). In this paper proposed method are implemented and
evaluated by using MATLAB software. Next in this paper proposed method has employee
two processes: one is POC based phase matching components of two same images and two
dierent images. Another one is to perform BLPOC based phase matching components of
two same images and two dierent images. This issue has been showed in Figure 4 and 5.
The Figure 5 shows POC level estimated at 0.64, it express high matching trial has achieved.
Uniformly Figure 6 shows BLPOC level estimated at 0.14, it express high matching trial
has achieved. It shows better performance than other modalities that are previously existed.
4.1. PERFORMANCE EVALUATION OF HUMAN EAR RECOGNITION
The Receiver Operating Characteristic (ROC) curve evaluates the performance of the
biometrics-based authentication system, which constructs with the help of False Reject
Rate (FRR) and False Accept Rate (FAR) in various thresholds resting on matching score.
Primarily, estimate the FRR in support of all the probable combinations number of authentic
attempts. Similarly, estimate the FAR in support of all the probable combinations number
of imposter attempts shown in Figure 7. In Figure 8, the ROC curve belongs to BLPOC; it
demonstrates this algorithm is suitable for recognizing human ear images. Hence, proposed
algorithm consider overall and conned deformation of human ear images together to
calculate the scores of matching among the human ear images. Table 2 shows the EER [%]
and distance (d’) values of Ear identication.
(16)
Table 2. Equal error rate and distance (d’) of the human ear recognition.
Proposed algorithm Equal error rate (%) Distance (d’)
BLPOC 0.86 2.145
152 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Figure 7. EER rate of proposed system.
Figure 8. ROC curves for BLPOC algorithm.
Where, the mean value of genuine and imposter matching scores, respectively to an
authentic and fraud identication, and the standard deviation of genuine and imposter
matching scores, respectively to an authentic and fraud identication. The performances of
a proposed method are able to assess in terms of identication accuracy, which is specied
in the equation (17).
(17)
153 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Figure 9. Accuracy of BLPOC algorithm.
The Figure 9 shows the accuracy of BLPOC algorithm. Where, FRR is the false
rejection rate at which an authentic person is wrongly rejected as an imposter and
FAR is the false acceptance rate, which denotes the rate of an imposter is wrongly accepted
as an authentic person.
5. CONCLUSION
This paper intends a human ear identication using phase based function matching
algorithm. The proposed BLPOC algorithm of Human Ear Recognition makes it possible
to align ear images, correctly evaluated similarity between them and obtained the reliable
matching score. The experimental results reveal that the intended method has achieved
the well again Human Ear recognition performance and robustness over other previous
methods. Extensively experiments were tested on the Ear database IIT Delhi. The proposed
method of human ear recognition has performed the nest verication results on the Ear
database IIT Delhi, with the equal error rate 0.86%.
ACKNOWLEDGEMENT
I thank the department of Electronics and Communication Engineering of Kalasalingam
University, (Kalasalingam Academy of Research and Education), Tamil Nadu, India for
permitting to use the computational facilities available in Signal Processing and VLSI
Design laboratory which was setup with the support of the Department of Science and
Technology (DST).
154 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
REFERENCES
Aoyama, S., Ito, K., & Aoki, T. (2014). A nger-knuckle-print recognition algorithm
using phase-based local block matching. Information Sciences, 268, 53-64. https://doi.
org/10.1016/j.ins.2013.08.025
Bustard, J. D., & Nixon, M. S. (2010). Toward unconstrained ear recognition from two-
dimensional images. IEEE transactions on systems, man, and cybernetics-Part A: Systems and
Humans, 40(3), 486-494. https://doi.org/10.1109/TSMCA.2010.2041652
IIT Delhi Ear Database. (2007). www.iitdelhi/biometric/ear.html
Ito, K., Nakajima, H., Kobayashi, K., Aoki, T., & Higuchi, T. (2004). A ngerprint
matching algorithm using phase-only correlation. IEICE transactions on fundamentals
of electronics, communications and computer sciences, 87(3), 682-691. http://www.aoki.ecei.
tohoku.ac.jp/research/docs/e87-a_3_682.pdf
Jain, A.K., Ross, A., & Pankanti, S. (2006). Biometrics: a tool for information security.
IEEE transactions on information forensics and security, 1(2), 125-143. https://doi.
org/10.1109/TIFS.2006.873653
Mahri, N., Suandi, S. A., & Rosdi, B. A. (2010, August). Finger vein recognition algorithm
using phase only correlation. In 2010 International Workshop on Emerging Techniques
and Challenges for Hand-Based Biometrics (pp. 1-6). IEEE. https://doi.org/10.1109/
ETCHB.2010.5559283
Malathy, C., Annapurani, K., & Sadiq, A. K. (2013). Analysis of fusion methods for
ear biometrics. International Journal of Advanced Research in Computer Science and software
Engineering, 3(1), 181-185.
Miyazawa, K., Ito, K., Aoki, T., Kobayashi, K., & Katsumata, A. (2006, October).
An iris recognition system using phase-based image matching. In 2006 International
Conference on Image Processing (pp. 325-328). IEEE. https://doi.org/10.1109/
ICIP.2006.313159
155 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Miyazawa, K., Ito, K., Aoki, T., Kobayashi, K., & Nakajima, H. (2005, September).
An ecient iris recognition algorithm using phase-based image matching. In IEEE
international conference on image processing 2005, (Vol. 2, pp. II-49). IEEE. doi: https://doi.
org/10.1109/ICIP.2005.1529988
Nanni, L., & Lumini, A. (2007). A multi-matcher for ear authentication. Pattern Recognition
Letters, 28(16), 2219-2226. https://doi.org/10.1016/j.patrec.2007.07.004
Rutty, G. N., Abbas, A., & Crossling, D. (2005). Could earprint identication be
computerised? An illustrated proof of concept paper. International Journal of Legal
Medicine, 119(6), 335-343. https://doi.org/10.1007/s00414-005-0527-y
Takita, K., Muquit, M. A., Aoki, T., & Higuchi, T. (2004). A sub-pixel correspondence
search technique for computer vision applications. IEICE transactions on fundamentals of
electronics, communications and computer sciences, 87(8), 1913-1923. http://www.aoki.ecei.
tohoku.ac.jp/research/docs/e87-a_8_1913.pdf
Yazdanpanah, A. P., & Faez, K. (2010). Ear Recognition Using Bi-Orthogonal and
Gabor Wavelet-Based Region Covariance Matrices. Applied Articial Intelligence, 24(9),
863-879. https://doi.org/10.1080/08839514.2010.514228
Yuan, L., Wang, Z. H., & Mu, Z. C. (2010, April). Ear recognition under partial
occlusion based on neighborhood preserving embedding. In Biometric Technology for
Human Identication VII (Vol. 7667, p. 76670Y). International Society for Optics and Photonics.
Zhang, L., Zhang, L., & Zhang, D. (2009, September). Finger-knuckle-print verication
based on band-limited phase-only correlation. In International Conference on Computer
Analysis of Images and Patterns (pp. 141-148). Springer, Berlin, Heidelberg.
Zhang, L., Zhang, L., Zhang, D., & Guo, Z. (2012). Phase congruency induced local
features for nger-knuckle-print recognition. Pattern Recognition, 45(7), 2522-2531.
https://doi.org/10.1016/j.patcog.2012.01.017
Zhang, L., Zhang, L., Zhang, D., & Zhu, H. (2011). Ensemble of local and global
information for nger–knuckle-print recognition. Pattern recognition, 44(9), 1990-1998.
https://doi.org/10.1016/j.patcog.2010.06.007
156 http://doi.org/10.17993/3ctecno.2020.specialissue4.141-157
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Marzo 2020
Zhu, Y. H., Jia, W., & Liu, L. F. (2009, September). Palmprint recognition using band-
limited phase-only correlation and dierent representations. In International Conference
on Intelligent Computing (pp. 270-277). Springer, Berlin, Heidelberg.
