20II International Conference on Computer Science and Network Technology

Personal Identification Using Computerized Human

Ear Recognition System

Anam Tariq*, M. Almas Anjumt, M. Usman Akram:�

Department of Computer Engineering,
National University of Sciences and Technology,
College of E&ME,
Rawalpindi, Pakistan.
anam.tariq86@[Link]*, almasanjum@[Link], usmakram@[Link]:�

Abstract-Biometric authentication for personal identification is

very popular now a days. Human ear recognition system is a new
technology in this field. The change of appearance with the
expression was a maj or problem in face biometrics but in case of
ear biometrics the shape and appearance is fixed. That is why it
is advantageous to use it for personal identification. In this paper,
we have proposed a new approach for an automated system for
human ear identification. Our proposed method consists of three
stages. In the first stage, preprocessing of ear image is done for its
contrast enhancement and size normalization. In the second stage,
features are extracted through Haar wavelets followed by ear
identification using fast normalized cross correlation in the third
stage. The proposed method is applied on USTB ear image
database and lIT Delhi ear image database. Experimental results
show that our proposed system achieves an average accuracy of
97.2% and 95.2% on these databases respectively.

Keywords- ear recognition; haar wavelet; fast normalized cross

correlation; automated system, biometric security Figure I. Anatomy of Human Ear

There are many approaches in the literature for an

I. INTRODUCTION
automated ear recognition system. Alfred Iannarelli [4] was a
Biometrics deals with identification of individuals on the basis pioneer in this field. He classified the ear into eight parts on
of their physiological and behavioral characteristics. Biometric the basis of 12 measurements. But this method was not
technology is gaining popularity in this modem era for the suitable because of the difficulty of localization of anatomical
purpose of security and other applications. points [4]. Burge and Burger [5] proposed that each
The new feature in biometrics is human ear which is individual's ear is modeled as a graph of adjacency that can be
becoming popular. It has several advantages over other built using Voronoi diagram of curve segments. For ear
biometric technologies such as iris, fmgerprints, face and recognition, they have proposed graph matching algorithm
retinal scans. Ear is large as compared to iris and fmgerprint which was suitable only for passive identification. Moreno
and unlike them, the image acquisition of human ear is very also proposed a new method for ear recognition based on outer
easy as it can be captured from a distance without the ear contour's feature points and information that can be
cooperation of individual [I]. Human ear contains rich and obtained from shape and wrinkles present in the ear [6]. In [1],
stable features and it is more reliable than face as the structure Hurley proposed an approach based on force field transform in
of ear is not subject to change with the age and facial which the ear image is considered as an array of Gaussian
expressions. The anatomy of human ear is given in figure I. It attractors that act as a source of force field. The ear is
shows the standard features of human ear. It has been found described by small channels by using these directional
that no two ears are exactly the same even that of identical properties of force field. The high order moment invariant
twins [2] [3]. Therefore it appears that ear biometrics is a good based technique was used by Wang to extract ear features [7].
solution for computerized human identification and Mu presented a shape and structural feature based ear
verification systems. The major application of this technology recognition method for the identification of persons. This edge
is crime investigation. Ear features have been used for many based ear recognition method includes edge detection, ear
years in the forensic sciences for recognition. description, feature extraction that is followed by ear

978-1-4577-1587-7/111$26.00 ©20111EEE 50 December 24-26,2011

recognition based on the feature vector which has properties of A. Preprocessing
both outer and inner ear (shape and structure) [8]. The Preprocessing of ear image is the first stage in our proposed
standard PCA algorithm was used by Chang for ear system. Preprocessing is done to segment out the ear image
recognition giving a multimodal approach for ear and face from the rest of the head portion of a person. Also, size
biometric recognition [3]. Yazdanpanah et al. proposed an ear normalization and ear image enhancement is the requirement
recognition system using bi-orthogonal and Gabor wavelet of our proposed system before feature extraction. Images with
based region covariance matrices approach. They achieved ear rings, other artifacts and occluded with hairs have not been
good average accuracy on USTB database [9]. Daramola et al. processed in this proposed technique. The main steps of
proposed an automatic ear recognition system using energy preprocessing stage are as below:
edge density feature and back propagation neural networks • Image is cropped using row and column wise mean based
[10]. scanning from the head image of the person.
This paper presents the novel technique for automated • The cropped images are of different sizes. So size
human ear recognition system. There are three steps of the
normalization of cropped ear images is done to convert
proposed algorithm. In the first step preprocessing is applied
them to a fixed size of 160x120 pixels.
on the ear image which includes its cropping, size
• The resized images are converted to grayscale.
normalization and contrast enhancement. In the second step,
• The contrast enhancement of grayscale image is done using
features of ear are extracted using Haar wavelet. In the third
contrast limited adaptive histogram equalization [11]. The
step of proposed system feature matching is done using fast
main importance of this method is to define a point
normalized cross correlation (NCC) which gives good results
transformation within a local large template or window
for template matching.
assuming that the intensity value within that window is a
Rest of the paper is organized as follows. In section II
stoical representation of local distribution of intensity
proposed ear recognition technology is discussed in detail
value within a whole ear image. Consider a running
explaining every step of the system. Experimental results are
suimage W(x,y) of NxN pixels centered on a pixel P(i,j),
discussed in section III and conclusion is given in section IV.
the image is fitered with another subimage P(x,y) of NxN
pixels according to equation (1):
II. PROPOSED SYSTEM

Figure 2 shows the systematic overview of the human ear

recognition technology. In the first step image is acquired Pn = 255
[ [<l>w(p) - <l>w(min)] 1 (1)
which is then cropped and enhanced in preprocessing stage. In
[<l>w(max) - <l>w(min)]
feature extraction stage, haar wavelet is applied to get the main
features of the ear. In the matching stage, fast normalized cross where <Pw is defined in equation (2):
correlation technique is used for the recognition of human ear.
Each of the components in this figure is described in detail in
this section. (2)

Image Acquisition
Preprocessing and min and max are the minimum and maximum intensity
values in the whole image respectively. Jlw and O"w are
the local window mean and standard deviation respectively
Haar
and are defmed by equation (3) and equation (4) which are
Wavelets as follows:
Fast
1
I
Ear Matching Fcamrc Extraction 14--.....
j..lw = -2 p(i, j) (3)
N (i,j)E(k,l)

o"w
1
I (p(i, j) - Jlw)2 (4)
N2
= .

(/,j)E(k,l)

Figure 2. Human Ear Recognition Technology

This reduces the noise that may be present in the image.
Hence the image becomes clearer after applying this
method. Figure 3 shows the preprocessing stage of our
proposed system.

51
 a)

Figure 3. Preprocessing: a)Original Image,b)Cropped Image, c)Resized

Image,d)Contrast Enhanced Image Figure 4. Second level Haar wavelet decomposition of preprocessed ear
image

B. Feature Extraction
After the preprocessing module, the next stage of human ear C. Ear Matching
recognition system is feature extraction. Feature extraction is After extraction of required features from the ear image,
done using Haar wavelet transform [12]. Haar wavelet is the matching is done using fast normalized cross correlation. The
fIrst known wavelet proposed by Alfred Haar in 1909. Let basic expression for Nee is given in equation (8) [13].
'P:R---+R, the mother Haar wavelet function is defIned by

L x,y (I(x,y)-l",J(t(x-u,y-v)- T)
equation (S):
1 r = --;================= (8)
l,for tE[O,-), " (t(x-u,y-v)- -t )2
2 �x,y (I(x,y)- IlI,J2 �
",,, x,y
1
\{f(t)= -l,for tE[- ,l), (S)
2 Where Ill,v is the mean value of I(x,y) within the area of
0, otherwise. template t shifted to (u,v). The simple Nee based technique
suffers from many problems since it is not invariant with
respect to imaging scale, rotation, and perspective distortions.
Hence fast Nee is used by taking all the features from Haar
and we can generate any Haar function using equation (6) wavelet decomposition in the second stage [13]. A Haar
wavelet coeffIcient based feature vector is computed for a test
\{f/ (t)= .J2i\{f(2i t -i) (6)
image and is correlated with the feature vectors saved in
database. Fast Nee for ear recognition is very signifIcant as it
requires less time for person identifIcation.
Where i=0,1, ....,2i-l and j = 0,1, ....,log2N-1. Now two
dimensional Haar wavelet transform can be computed using III. EXPERIMENTAL RESULTS
equation (7) We have tested our proposed algorithm on University of
Science and Technology Beijing (USTB) ear image database
N-IN-l
[14] and Indian Institute of Technology (lIT) Delhi ear image
S(l,m)= LIJ(x,y)H(l,x)H(m,y) (7) database [IS]. USTB ear image database 1 consists of 180
x=o y=o
right ear images, three images per person (60 persons). These
images are 8 bit gray scale and under different lighting
Where I(x,y) is a NxN subimage and H(2l + i,t)= \{f/ (t) . conditions. USTB ear image database 2 consists of total 308
right ear images, 4 images per person (77 persons). Four
For the proposed system, we have used Haar wavelet of level
images per person constitutes of one profIle image, two
two on the image for its decomposition. The desired features
images with angle variation and one image with illumination
of the ear are extracted using level two which are then used in
variation. Each image is 24-bit true color image and 300x400
the ear matching stage for the recognition of human ear.
pixels. The lIT Delhi ear image database is acquired from 12S
Figure 4 shows the feature extraction stage of our proposed
different subjects and each subject has at least 3 ear images.
system.

52
The resolution of these ear images is 272x204 pixels. Figure 5 No. of Recognized Unrecognized Recognition

shows the sample ear images from both these databases.

Classifiers subjects Ear Images Ear Images Rate (%)
Table I shows the recognition rate of our method on these Back

databases. The average accuracy on USTB ear database will propogation

neural 262 250 12 95.41%
be 97.2% while 95.2% on lIT ear image database. Results
network
show that recognition rate on USTB database is good. Table II [10]
shows the recognition rate of different matching approaches of Fast
ear recognition technology. Fast NCC performs well among normalized
262 252 10 96.18%
cross
several different classifiers in less time. Table III shows the
correlation
average execution time of different stages of our algorithm.
The experiments were carried out on an hp dv-6 workstation
with core i-5 (2.24 GHz) and with 4GB RAM. MATLAB 7.8 TABLE III. AVERAGE PROCESSING TIME OF DIFFERENT STAGES OF OUR
(R2009a) revised version in windows (64-bits) platform was METHOD
used for the performance evaluation. Feature Ear
Preprocessing
Database Extraction Recognition
(secs)
(secs) (secs)

USTB DB I 0.11 0.03 0.18

USTB DB 2 0.23 0.05 0.21

liT Delhi
0.33 0.04 0.23
Database

These results show the soundness of our algorithm and proved

that the proposed system not only provides reliability of the
results but also very efficient. Also, our ear matching approach
is better than previous methods.

IV. CONCLUSION

In this paper, a novel approach is presented for an automated

human ear recognition system. The approach consists of three
stages such as preprocessing, Haar wavelet based feature
extraction and finally fast NCC based ear feature matching.
Experimental results on the two databases show that our
proposed technique is good and effective as it gives better
a) b) results than other matching algorithms. The recognition rate is
Figure 5. Sample Ear Images: Column a) USTB Ear Database,Column b) also good than previous approaches for ear recognition. As for
liT Delhi Ear Database future work of this research, ear localization property should
be integrated in the preprocessing module.

TABLE I. RECOGNITION RATE OF PROPOSED METHOD ON DIFFERENT

DATABASES
ACKNOWLEDGMENT

No. of Recognized Unrecognized Recognition

The authors would like to thank University of Science and
Database subjects Ear Images Ear Images Rate (%)
Technology Beijing (USTB) and Indian Institute of
USTB Technology (lIT) Delhi for providing their databases on
60 59 I 98.33%
DB 1 request for carrying out this research work.
USTB
77 74 3 96.1%
DB 2
REFERENCES
liT Ear
125 119 6 95.2% [I]. DJ. Hurley, M.S. Nixon, 1. N. Carter, "Force field feature extraction
Database
for ear biometrics",Computer vision and image understanding,Vol. 98,
No. 3,pp.491-512,2005.
[2]. B. Victor, K. Bowyer, and S. Sarkar, "An Evaluation of face and Ear
TABLE II. RECOGNITION RATE OF DIFFERENT CLASSIFIERS Biometric", 16th International conference of Pattern Recognition,
pp.429-432, 2002.
No. of Recognized Unrecognized Recognition
[3]. K. Chang,[Link],and V. Barnabas, "Comparison and Combination
Classifiers subjects Ear Images Ear Images Rate (%)
of Ear and Face Image in Appearance-Based Biometrics", IEEE
Euclidean Transactions on Pattern Analysis and Machine Intelligence,Vol. 25,pp.
distance 262 238 24 90.83% 1160-1165,2003.
[16]

53
[4]. A. lannarelli, "Ear Identification," Forensic Identification Series, & image processing and network security, TJVIPNS-IJENS, Vol. 11,
Paramount Publishing Company,Fremont,California,1989. No.1, pp.28-32,2011
[5]. M. Burge and W. Burger, "Ear Biometrics in Computer Vision," 15th [11]. K. Zuiderveld, "Contrast Limited Adaptive Histograph
International Conference of Pattern Recognition,pp.822-826,2000. Equalization. " Graphic Gems IV. San Diego: Academic Press
[6]. B. Moreno, A. Aanchez and 1. F. Velez, "Use Outer Ear Images for Professional,pp. 474--485,1994.
d
Personal Identification in Security Applications," 33c International [12]. C. Burrus, R. Gopinath, H. Guo, "Introduction to Wavelets and
Carnahan Conference,Madrid,pp. 469-476, 1999. Wavelet Transforms", Prentice Hall,New Jersy, 1998
[7]. W. Zhongli, M. Zhichun, W. Xiuyan, M. Hongtao, "Ear recognition [13]. 1. P. Lewis, "Fast Template Matching", Vision Interface, pp. 120-123,
based on Moment Invariant Matching ", Pattern Recognition and 1995.
Artificial Intelligence,Vol 17,No. 4,pp. 502-505,2004. [14]. L. Yuan, M. Zhichun, X. Zhengguang, "Using ear biometrics for
[8]. M. Zhichun, Y. Li, X. Zhengguang, "Shape and Structural Feature personal recognition", Advances in Biometric Person Authentication,
Based Ear Recognition," Advances in Biometric Person Authentication, LNCS,Vol. 3781,pp.221-228,2005.
Guangzhou,China,pp. 663-670, 2004. [15]. A. Kumar and C. Wu, "Automated human identification using ear
[9]. A. P. Yazdanpanah and K. Faez, "Ear Recognition using biorthogonal imaging," Pattern Recognition,2011. (in press)
and Gabor wavelet based region covariance matrices, Applied [16]. M. Ali, M. Y. Javed, A. Basit, "Ear recognition using wavelets',
Artificial Intelligence, Vol. 24, No.9, pp.863-879,2010 Proceedings of Image and Vision computing New Zealand, pp.83-86,
[10]. S. A. Daramola,O. D. Oluwaninyo, "Automatic ear recognition system Dec.2007.
using back propagation neural network" ,International Journal of video

54