Scribd
Upload
166 views34 pages

0928 - 1600 - Mr15 - Ross Ear PDF

Uploaded by

Lavanya Kalapala
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
166 views34 pages

0928 - 1600 - Mr15 - Ross Ear PDF

Uploaded by

Lavanya Kalapala
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 34

Advances in Ear

Biometrics
Arun Ross
Robert C. Byrd Associate Professor
West Virginia University
arun.ross@mail.wvu.edu
http://www.csee.wvu.edu/~ross

Acknowledgments
Ear Recognition: Was funded by the FBI Academy
through TSWG
Ear Classification: Was funded by FBI-BCOE
Collaborators:
Dr. Ayman Abaza and colleagues, WVHTF
Dr. Mohamed Abdel-Mottaleb, Univ of Miami
Ravindra Gadde, WVU

Ross 2011

Goal of this work


Categorizing ear into multiple classes
Benefits:
Faster matching
Subject exclusion
Forensic analysis

Ross 2011

INTRODUCTION

Ross 2011

Ear as a Biometric
The morphology of the outer
ear (pinna) is simple
compared to the rich texture of
the iris or the random distribution
of minutiae in fingerprints

The structure of the ear is


fairly stable, and robust to
change in facial expressions

From Hurley and Nixon, The Ear


as a Biometric, EUSIPCO 2007
Ross 2011

Alfred Iannerelli
Next to fingerprints, the external ear constitutes
the most unique design, characteristic features and
peculiarities for the purpose of identification. On no
other part of human body do we have flesh lines
with such a unique design
Reference: Ear Identification, Forensic Identification series by Alfred V. Iannarelli

IS THIS REALLY TRUE?

Ross 2011

Alphonse Bertillon
The ear, thanks to these multiple
small valleys and hills which
furrow across it, is the most
significant factor from the point of
view of identification.
"It is, in fact, almost impossible to
meet with two ears which are
identical in all their parts. . ."
IS THIS REALLY TRUE?
Reference: Alphonse Bertillon. Legal photograph, Gauthier-Villars, Paris 1890.
Ross 2011

External Anatomy of the Ear


1-Helix Rim
2-Lobule
3-Antihelix
4-Concha
5-Tragus
6-Antitragus
7-Crus of Helix
8-Triangular Fossa
9-Incisure Intertragica
Burge and Burger (1998)

For an interactive review of the ear anatomy see:


http://www.plasticsurgery4u.com/procedure_folder/otoplasty_anatomy.html

Ross 2011

External Anatomy of the Ear

Hurley et al, The Ear as a Biometric, in the Handbook of Biometrics, 2007

Ross 2011

Are Ears Unique? Part I


Through 38 years of research and application
in earology, the author* has found that in
literally thousands of ears that were examined
by visual means, photographs, ear prints, and
latent ear print impressions, no two ears were
found to be identical not even the ears of any
one individual..
This uniqueness held true in cases of identical
and fraternal twins, triplets, and quadruplets

* A. Iannarelli, Ear Identification, Forensic Identification Series, Paramont Publishing Company,


Fremont, California, 1989.
Ross 2011

Are Ears Unique? Part II


A study done in California compared over 10,000
ears drawn from a randomly selected sample
no two ears were indistinguishable

Another study examined the ear structures of


fraternal and identical twins while their ear
structures were similar, they were still clearly
distinguishable

Imhofer (1906) studied a set of 500 ears and

noted that he could clearly distinguish between


ears based on only 4 features

*Hoogstrate et al Ear Identification Based on Surveillance Cameras Images, May 31, 2000
Ross 2011

Automated Feature Sets

Note: Images taken from original publications of various authors

Ross 2011

More Research Needed


More research is needed to examine the validity
of uniqueness that has been claimed in the past

However, potential of ear as a soft biometric cue


Side-profile face images (mugshots)
Exclusion of individuals
Classification of faces
Fusion with face or other evidence

Ross 2011

Ear Signature
An ear signature is generated

by probing the ear with a


sound signal which is reflected
and picked up by a small
microphone

The shape of the pinna and

the ear canal determine the


acoustic transfer function
which forms the basis of the
signature
*Ackerman et al., Acoustic Ear Recognition for Person Identification, 2005

Ross 2011

EAR DETECTION

Ross 2011

AdaBoost: Fast Feature Selection

A feature selection method based on Forward


Feature Selection (FFS)
The most time consuming component is moved out
of the loop
Original Adaboost needs to be trained T times,
where T is the number of features in the AdaBoost
ensemble
FFS requires only 1/T training time as that of
AdaBoost
Summary: The key intuition in FFS is
precomputing: the results of trained weak
classifiers are stored and re-used

Fast Asymmetric Learning for Cascade Face Detection, Jianxin Wu, S. Charles Brubaker, Matthew D. Mullin, and
James M. Rehg, IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(3), 2008: pp. 369-382.
Ross 2011

WVU Database

Face images of an user which vary from side profile


to frontal profile in small increments
120 subjects with 10 images per subject
The original image are of size 480x640 and the
cropped ear images have variable sizes
Typical average cropped ear is of size 128x94:
these images are used for training, so images
scaled to 24x16

Ross 2011

UND Database

Side profile images of a subject taken at arbitrary


angles and under different lighting conditions
114 subjects with arbitrary number of images per
subject
Size of each image is 1600x1200
These images are used for testing, so the original
images are directly input to the detector without
any scaling

Ross 2011

Magna Database

Images taken from two different sensors and five


different profiles
3360 subjects with 8 images per subject with
different profiles: Left and Right profile images are
used in our experiments
The size of each image is 1200x1600
These images are used for testing: so the original
images are directly used without any scaling

Sample images from a single user

Ross 2011

Ear Detection: Results on MAGNA

Good: If the structure of the ear is clearly visible


Partial: If the structure of the ear is cloudy due to
some obstacles
Bad: If the ear is not completely visible
Type

FAR

FRR

Good

8.6% (178 of 2069)

9.3% (194 of 2069)

Partial

11.3% (47 out of 414)

18.1% (75 out of 414)

Ross 2011

CLASSIFICATION

Ross 2011

Ear Classification

Automatic classification of ear image into one of


several pre-determined classes

Reduces the search time by searching only a subset


of images
Ross 2011

Our Approach

Lobule, helix and anti-helix represent a


predominant portion of the ear
Classification in 4-Levels:
Level-1 based on color
Level-2 based on the contour of helix and lobule
Level-3 based on contour of anti-helix and antitragus
Level-4 based on ear lobule

Ross 2011

Why Level-based?

Due to missing features in certain poses we need a


level based approach

Pose-1:Features at all levels are present Pose-2: Level-3 & 4 features are absent
Ross 2011

UND Database

Sample images from the database

Ross 2011

Level-1: Color

Classification based on color is challenging because


of its sensitivity to illumination variation and
change in chromaticity

Classifier can be designed based on the changes in


chromaticity at different lighting conditions for
different skin tones and for different camera
calibration (assuming there is sufficient training
data)

*B. Martinkauppi, M. Soriano, M. Pietikainen, Detection of skincolor under changing illumination: a


comparative study, 12th Interenational conference on Image Analysis and Processing, 2003

Ross 2011

Level-2: Helix + Lobule

The outer contour is approximated using an ovoid


model
Scale invariant parameters such as ratio of minor
axis to major axis of the ear and the extent of
distortion () from an ellipse are measured
= 0.3

= 0.6

* Luis Alvarez, Esther Gonzalez and Luis Mazorra. Fitting Ear Contour Using Ovoid Model, 2005

Ross 2011

Level-2: Helix + Lobule

The parameter shows variation across subjects.


However, it is not pose invariant
A hit rate of 95% possible for a two-class problem
even if each subject exhibits different profiles
(side, back, middle)

Ross 2011

Level-3: Anti-Helix + Anti-tragus

Inner contour along with Concha


region (yellow curve)

Inner contour along with the upper


part of Triangular Fossa (white
curve)

A snake algorithm used to estimate


the shape of the target region

The contour of the upper part of


triangular fossa is not properly
estimated because of shadows

Ross 2011

Level-4: Lobules

Human earlobes may be free or detached


(hanging free from the head) or attached (joined
to the head)

Out of 114 subjects in the UND database, 81


subjects possess detached lobule and 33 have
attached lobule

The attachment or detachment of the lobule can


only be observed from certain profiles of the ear

Ross 2011

Manual Classification

Iannarellis classification:
round
rectangle
triangle
oval

Ear Identification (Forensic Identification Series) by Alfred V. Iannarelli

Ross 2011

Manual Classification: Results


Class
Round

Good

Partial

Bad

71

15

1356

277

523

119

Rectangular

62

Undetermined

57

304

325

Oval
Triangular

Total

2069 (66.4%)

718 (23%)

325 (10.4%)

Ross 2011

Ongoing Work
Relating the helix-based classification to
Iannarellis system using shape
Using position of the ear relative to the eye brow
Automating the process of detecting landmarks
and contours
Modifying the snake algorithm used for estimating
the shape of triangular fossa
Finding new parameters to represent the shape of
concha
Ross 2011

Advances in Ear
Biometrics
Arun Ross
Robert C. Byrd Associate Professor
West Virginia University
arun.ross@mail.wvu.edu
http://www.csee.wvu.edu/~ross

You might also like