Biometric sensors

• A biometric sensor is a transducer that

changes a biometric traits of a person into an
electrical signal.
• Every biometric device requires at least
one type of sensor
• The working of a sensor may be based on the
measurement of one or more physical
quantities like light intensity, capacitance,
temperature, image, etc.
Biometric sensors
 Types of Biometric Sensors
• Biometrics is broadly divided into three
categories as follows
– Biological biometrics
– Morphological biometrics
– Behavioral biometrics
 Types of Biometric Sensors
• Biological biometrics sensors involve biological
measurements at the genetic or molecular level,
and DNA sequencing is a biological biometric
system.
• It requires the sampling of DNA from blood or
bodily fluids.
• This cannot be used for security or
authentication systems but has other practical
applications like DNA matching, genetic diseases,
and microbiological studies.
 Types of Biometric Sensors
• Morphological biometrics sensors involve the
measurement of physical traits and body
structures.
• The security and authentication systems are
usually based on one or more morphological
biometrics.
• This includes fingerprint mapping, iris scan, face
recognition, finger geometry recognition, vein
recognition, hand geometry, ear recognition, and
odour recognition.
 Types of Biometric Sensors
• Behavioral biometrics sensors involve the
measurement of behavioral identifiers unique to
a person.
• These systems are not common but are reserved
for special applications.
• This includes signature recognition, voice
biometrics, gait biometrics, keystroke
recognition, and gesture recognition.
 Data Collection subsystem
• Also called Data Acquisition
• Comprises input device or sensor that reads the
biometric information from the user
• Converts biometric information into a suitable
form for processing by the remainder of the
biometric system
• Examples: video camera, fingerprint scanner,
digital tablet, microphone, etc.

7
 Challenges in Data Collection
• The biometric features may change
• The presentation of the biometric feature at
the sensor may change
• The performance of the sensor itself may
change
• The surrounding environmental conditions
may change

8
 Image Preprocessing
• Preprocessing involves steps to make image
suitable for feature extraction and matching
• Most of the engineers spend a good amount
of time in data pre-processing before building
the model.
• The aim of pre-processing is an improvement
of the image data that suppresses undesired
distortions or enhances some image features
relevant for further processing and analysis
task.
9
 Image Preprocessing
The steps to be taken are :
• Read image
• Image Enhancement
– Remove noise
– Segmentation
– Removing Specular Highlights
– Region of Interest Detection and so on…
• Morphology(smoothing edges)

10
 Image Enhancement
• Image enhancement refers to the process of
highlighting certain information of an image, as
well as removing any unnecessary information
according to specific needs.
• The tools used for image enhancement include
many different kinds of software such as filters,
image editors and other tools for changing various
properties of an entire image or parts of an image.
• The principal objective of Image Enhancement is to
modify attributes of an image to make it more
suitable for a given task and a specific observer

11
 Image Enhancement

• The greatest difficulty in image enhancement is

quantifying the criterion for enhancement and,
therefore, a large number of image
enhancement techniques are empirical and
require interactive procedures to obtain
satisfactory results.
• Image enhancement methods can be based on
either spatial or frequency domain techniques

12
Image Enhancement (Spatial Domain)

13
Image Enhancement (Frequency Domain)
Frequency Domain

14
 Image Enhancement Techniques

Image Enhancement by Point Processing

1. Simple Intensity Transformation
a. Image Negatives
b. Contrast Stretching
c. Compression
d. Gray-level Slicing
2. Histogram Processing
a. Histogram Equalization
b. Histogram Specification
c. Local Enhancement

15
Image Enhancement by Point Processing

These processing methods are based only on the

intensity of single pixels.
1. Simple intensity transformation:
a. Image Negatives
• Negatives of digital images are useful in
numerous applications, such as displaying
medical images and photographing a screen
with monochrome positive film with the idea of
using the resulting negatives as normal slides

16
Image Enhancement by Point Processing

17
Image Enhancement by Point Processing

18
Image Enhancement by Point Processing

19
Image Enhancement by Point Processing

20
 Image Enhancement by Point Processing

C) Compression of dynamic range

• Sometimes the dynamic range of a processed
image far exceeds the capability of the display
device, in which case only the brightest parts of
the images are visible on the display screen.
• An effective way to compress the dynamic range
of pixel values is to perform the following
intensity transformation function:

21
Image Enhancement by Point Processing

22
 Image Enhancement by Point Processing
(d) Gray-level slicing
• Highlighting a specific range of gray levels in an image often is
desired.
• Applications include enhancing features such as masses of
water in satellite imagery and enhancing flaws in x-ray images
• Example: A transformation function that highlights a range
[A,B] of intensities while diminishing all others to a constant.

23
Image Enhancement by Point Processing

24
 Image Enhancement by Point Processing
2 Histogram processing:

25
 Image Enhancement by Point Processing
2 Histogram processing:

26
Image Enhancement by Point Processing

27
Image Enhancement by Point Processing

28
Image Enhancement by Point Processing

29
Image Enhancement by Point Processing

30
Image Enhancement by Point Processing

31
Image Enhancement by Point Processing

32
Image Enhancement by Point Processing

33
Image Enhancement by Point Processing

34
Image Enhancement by Point Processing

35
Image Enhancement by Point Processing

36
Image Enhancement by Point Processing

37
Image Enhancement by Point Processing

38
Enhancement in the frequency domain:

39
Enhancement in the frequency domain:

40
Enhancement in the frequency domain:

41
 Removing Specular Highlights
• Image contains specular highlights due to source of
light
• Such highlights can be removed in 2 stages
– Adaptive Thresholding - Adaptive thresholding is
the method where the threshold value is
calculated for smaller regions and therefore, there
will be different threshold values for different
regions.
– Hole Filling - The hole-filling method is to fill in
the holes in the depth image captured from the
sensors used for rendering.

9/3/2010 Phalguni Gupta, Biometrics Lab, IIT Kanpur 42

 thresholded image

Input Image Binarized Image Image after hole filling

with holes

Removal of Specular Highlights from Iris

9/3/2010 Phalguni Gupta, Biometrics Lab, IIT Kanpur 43

 Region of Interest Detection
• A region of interest (ROI) is a portion of an image that
you want to filter or operate on in some way.
• You can represent an ROI as a binary mask image.
• In the mask image, pixels that belong to the ROI are set
to 1 and pixels outside the ROI are set to 0.
• The Matlab toolbox offers several options to specify ROIs
and create binary masks.
• It supports a set of objects that you can use to create
ROIs of many shapes, such circles, ellipses, polygons,
rectangles, and hand-drawn shapes.
• After you create the objects, you can modify their shape,
position, appearance, and behavior.
9/3/2010 Phalguni Gupta, Biometrics Lab, IIT Kanpur 44
 Palm print Extraction - ROI

Region of interest in Binary Image Region of Interest in Original Image

9/3/2010 Phalguni Gupta, Biometrics Lab, IIT Kanpur 45
 Image Restoration
• Image restoration is the operation of taking a
corrupt/noisy image and estimating the clean,
original image.
• Corruption may come in many forms such
as motion blur, noise and camera mis-focus.
• Image restoration is performed by reversing
the process that blurred the image and such is
performed by imaging a point source and use
the point source image, which is called the
Point Spread Function (PSF) to restore the
image information lost to the blurring process.
 Image Restoration
Image restoration attempts to restore images
that have been degraded
– Identify the degradation process and attempt to
reverse it
– Similar to image enhancement, but more
objective
Image degradations
 Image Restoration
• Image restoration is different from image
enhancement
• Image enhancement is designed to emphasize
features of the image that make the image more
pleasing to the observer,
• Image restoration is to produce realistic data from
a scientific point of view.
• The most straightforward and a conventional
technique for image restoration is deconvolution,
which is performed in the frequency domain
Image Restoration
Image Restoration
 Image Segmentation
• Goal of Segmentation to simplify and/or change the
representation of an image into something that is
more meaningful and easier to analyze.
• Image segmentation is a method in which a digital
image is broken down into various subgroups called
Image segments which helps in reducing the
complexity of the image to make further processing
or analysis of the image simpler.
• Segmentation is assigning labels to pixels.
• All picture elements or pixels belonging to the same
category have a common label assigned to them.
Image Segmentation 52
 Image Segmentation
• For example: Let’s take a problem where the picture
has to be provided as input for object detection.
• Rather than processing the whole image, the
detector can be inputted with a region selected by a
segmentation algorithm.
• This will prevent the detector from processing the
whole image thereby reducing inference time.

Image Segmentation 53
Approaches in Image Segmentation
• Similarity approach: This approach is based on
detecting similarity between image pixels to form a
segment, based on a threshold. ML algorithms like
clustering are based on this type of approach to
segment an image.
• Discontinuity approach: This approach relies on the
discontinuity of pixel intensity values of the image.
Line, Point, and Edge Detection techniques use this
type of approach for obtaining intermediate
segmentation results which can be later processed to
obtain the final segmented image

Image Segmentation 54
 Image Segmentation Techniques
1. Threshold Based Segmentation
2. Edge Based Segmentation
3. Region-Based Segmentation
4. Clustering Based Segmentation
5. Artificial Neural Network Based Segmentation

Image Segmentation 55
 Threshold based Segmentation
• Image thresholding segmentation is a simple form of
image segmentation.
• It is a way to create a binary or multi-color image
based on setting a threshold value on the pixel
intensity of the original image
• In this thresholding process, we will consider the
intensity histogram of all the pixels in the image.
• Then we will set a threshold to divide the image into
sections.

Image Segmentation 56
 Threshold based Segmentation
• For example, considering image pixels ranging from 0
to 255, we set a threshold of 60.
• So all the pixels with values less than or equal to 60
will be provided with a value of 0(black) and all the
pixels with a value greater than 60 will be provided
with a value of 255(white).
• But this threshold has to be perfectly set to segment
an image into an object and a background.

Image Segmentation 57
Threshold Based Segmentation

Image Segmentation 58
 Edge Based Segmentation
• Edge-based segmentation relies on edges found in an
image using various edge detection operators.
• These edges mark image locations of discontinuity in
gray levels, color, texture, etc.
• When we move from one region to another, the gray
level may change. So if we can find that discontinuity,
we can find that edge
• We can use various edge detectors like Sobel edge
operator, canny edge detector, Kirsch edge operator,
Prewitt edge operator, Robert’s edge operator, etc.
Image Segmentation 59
Edge Based Segmentation

Image Segmentation 60
 Region-Based Segmentation
• Region-based segmentation involves dividing an image
into regions with similar characteristics.
• The similarity between pixels can be in terms of intensity,
color, etc.
• Each region is a group of pixels, which the algorithm
locates via a seed point.
• Some predefined rules are set which have to be obeyed
by a pixel in order to be classified into similar pixel
regions. The preferred rule can be set as a threshold.
• Region-based segmentation methods are preferred over
edge-based segmentation methods in case of a noisy
image.
Image Segmentation 61
 Region-Based Segmentation
• Region-Based techniques are further classified into 2
types based on the approaches they follow.
– Region growing method
– Region splitting and merging method
• In Region growing method, we start with some pixel
as the seed pixel and then check the adjacent pixels.
• If the adjacent pixels abide by the predefined rules,
then that pixel is added to the region of the seed
pixel and the following process continues till there is
no similarity left.
• This method follows the bottom-up approach.
Image Segmentation 62
 Region-Based Segmentation
• In Region splitting, the whole image is first taken as a
single region.
• If the region does not follow the predefined rules,
then it is further divided into multiple regions
(usually 4 quadrants) and then the predefined rules
are carried out on those regions in order to decide
whether to further subdivide or to classify that as a
region.
• This process continues till there is no further division
of regions required i.e every region follows the
predefined rules.
Image Segmentation 63
 Region-Based Segmentation
• Usually, first region splitting is done on an image so as to
split an image into maximum regions, and then these
regions are merged in order to form a good segmented
image of the original image.
• In case of Region splitting, the following condition can be
checked in order to decide whether to subdivide a region
or not.
• If the absolute value of the difference of the maximum
and minimum pixel intensities in a region is less than or
equal to a threshold value decided by the user, then the
region does not require further splitting.
Image Segmentation 64
Cluster-Based Segmentation

Image Segmentation 65
 Cluster-Based Segmentation
• Clustering algorithms are unsupervised classification
algorithms that help identify hidden information in
images.
• One of the most dominant clustering-based
algorithms used for segmentation is KMeans
Clustering.
• This type of clustering can be used to make segments
in a colored image.
• The algorithm divides images into clusters of pixels
with similar characteristics, separating data
elements and grouping similar elements into clusters
Image Segmentation 66
 Cluster-Based Segmentation
KMeans Clustering
• Let’s imagine a 2-dimensional dataset for better
visualization.
• First, in the dataset, centroids (chosen by the user) are
first randomly initialized.
• Then the distance of all the points to all the clusters is
calculated and the point is assigned to the cluster with
the least distance.
• Then centroids of all the clusters are recalculated by
taking the mean of that cluster as the centroid.
• Then again data points are assigned to those clusters.
And the process continues till the algorithm converges to
a good solution.
Image Segmentation 67
Cluster-Based Segmentation

Image Segmentation 68
 Pattern classification and Extraction
• Pattern is everything around in this digital
world.
• A pattern can either be seen physically or it
can be observed mathematically by applying
algorithms.
• Example: The colors on the clothes, speech
pattern, etc. In computer science, a pattern is
represented using vector feature values.
 Pattern classification and Extraction
• What is Pattern Recognition?
Pattern recognition can be defined as the
classification of data based on knowledge already
gained or on statistical information extracted from
patterns and/or their representation.
• One of the important aspects of pattern recognition is
its application potential.
• Examples: Speech recognition, speaker identification,
multimedia document recognition (MDR), automatic
medical diagnosis.
 Pattern classification and Extraction
• In a typical pattern recognition application, the raw data is
processed and converted into a form that is amenable for a
machine to use.
• Pattern recognition involves the classification and cluster of
patterns.
• In classification, an appropriate class label is assigned to a
pattern based on an abstraction that is generated using a
set of training patterns or domain knowledge.
Classification is used in supervised learning.
• Clustering generated a partition of the data which helps
decision making, the specific decision-making activity of
interest to us.
• Clustering is used in unsupervised learning.
 Pattern classification and Extraction
• Pattern is defined as composite of features that are
characteristic of an individual.
• In classification, a pattern is a pair of variables {x,w}
where x is a collection of observations or features
(feature vector) and w is the concept behind the
observation (label).
• The quality of a feature vector is related to its ability
to discriminate examples from different classes
• Examples from the same class should have similar
feature values and while examples from different
classes having different feature values.
Pattern classification and Extraction

Fig: Characteristic (feature); a. the distinction between good and

poor features, and b. feature properties
 Pattern classification and Extraction
• The goal of a classifier is to partition feature space
into class-labeled decision regions.
• Borders between decision regions are called
decision boundaries

Classifier and decision boundaries

 Pattern classification and Extraction
• If the characteristics or attributes of a class are
known, individual objects might be identified as
belonging or not belonging to that class.
• The objects are assigned to classes by observing
patterns of distinguishing characteristics and
comparing them to a model member of each class.
• A pattern recognition system contains a sensor, a
preprocessing mechanism (segmentation), a feature
extraction mechanism (manual or automated), a
classification or description algorithm, and a set of
examples (training set) already classified or described
(post-processing)
Pattern classification and Extraction

A pattern recognition system

Steps in making decision
 Pattern classification and Extraction
Sensing
• The input to a pattern recognition system is often
some kind of a transducer, such as a camera or a
microphone array.
• The difficulty of the problem may well depend on
the characteristics and limitations of the
transducer- its bandwidth, resolution, sensitivity,
distortion, signal-to-noise ratio, latency, etc.
 Pattern classification and Extraction
Segmentation and Grouping
• We assumed that each object can be isolated, separate
from others and could easily be distinguished from the
Others.
• In practice, the object would often be overlapping, and
our system would have to determine where one ends and
the next begins-the individual patterns have to be
segmented.
• If we have already recognized the object then it would be
easier to segment their images.
• Segmentation is one of the deepest problems in pattern
recognition. Closely related to the problem of
segmentation is the problem of recognizing or grouping
together the various pans of a composite object.
 Pattern classification and Extraction
Feature Extraction
• The conceptual boundary between feature extraction and
classification proper is somewhat arbitrary: An ideal
feature extractor would yield a representation that makes
the job of the classifier trivial; conversely, an omnipotent
classifier would not need the help of a sophisticated
feature extractor The distinction is forced upon us for
practical rather than theoretical reasons.
• As with segmentation, the task of feature extraction is
much more problem- a domain-dependent and thus
requires knowledge of the domain
 Pattern classification and Extraction
Classification
• The task of the classifier component proper of a full
system is to use the feature vector provided by the
feature extractor to assign the object to a category.
• Because perfect classification performance is often
impossible, a more general task is to determine the
probability for each of the possible categories.
• The abstraction provided by the feature-vector
representation of the input data enables the
development of a largely domain-independent theory of
classification
• All non-trivial decision and pattern recognition problems
involve noise in some form.
 Pattern classification and Extraction
Post Processing
• The post-processor uses the output of the classifier to
decide on the recommended action.
• The simplest measure of classifier performance is the
classification error rate -the percentage of new patterns
that are assigned to the wrong category. Thus, it is
common to seek minimum-error-rate classification.
• We could also do better if we used multiple classifiers,
each classifier operating on different aspects of the input.
example
 Fingerprint
• Fingerprint Recognition and acquisition
• Fingerprint features, matching and synthesis
 Fingerprint
• The fingerprint is a physical biometric aspect.
• It is used to identify a person’s identity due to its
uniqueness where no two persons can share the
same fingerprint.
• Besides, a fingerprint is unchangeable with time
and can be easily recognized during the whole life
of the individual.
• The fingerprint is an impression or model of ribs
and valleys at the top of a person’s fingers.
 Fingerprint
• Fingerprint recognition is the automatic process
of comparing saved fingerprint pattern with the
input fingerprint to determine human characters.
• The fingerprint identification system is a cheap
but solid mechanism at the same time. Moreover,
it’s a simple way to identify humans speedily and
accurately
• Many applications applied fingerprint recognition
such as the military, judiciary, health, teaching,
civic serving, mobiles and laptop log-in, and many
more.
 Introduction
Fingerprint
Interleaved ridges
and valleys
Ridge width: 100μm-
300 μm
Ridge-valley cycle:
500 μm

87
 Levels of features
• Level 1 features(orientation field or ridge flow
and singular points)
• Level 2 feature(ridge skeleton)
• Level 3 features(ridge contour,pore,dot)
Three different levels of features
 Three different levels of features
• At the global level a fingerprint shows a
smoothed structure except in one or more
regions containing distinctive characteristics
called singularities which can be:
• delta (represented with the symbol ∆);
• loop (represented with the symbol ∩);
• whorl (represented with the symbol O)
 Fingerprint acquisition
• The first step in fingerprint recognition is image
acquisition - the process of capturing and digitizing the
fingerprint of an individual for further processing.
• The primary reason for the popularity of fingerprint
recognition is the availability of mature, convenient,
and low-cost sensors that can rapidly acquire the
fingerprint of an individual with minimum or no
intervention from a human operator.
• These compact fingerprint sensors have also been
embedded in many consumer devices such as laptops
and mobile phones.
 Fingerprint acquisition
• Finger Print Scanners
 Fingerprint Acquisition: Sensing Techniques
• Digital images of the fingerprints can be acquired using off-line or
on-line methods.
Off line Method:
• Off-line techniques generally do not produce the digital image
directly from the fingertip. Rather, the fingerprint is first
transferred to a substrate (e.g., paper) that is subsequently
digitized.
• For example, an inked fingerprint image, the most common form of
off-line capture, is acquired by first applying ink to the subject’s
fingertip and then rolling or pressing the finger on paper, thereby
creating an impression of the fingerprint ridges on paper.
• This kind of fingerprint is often called rolled fingerprint.
• A very important kind of off-line fingerprint image is the latent
fingerprint: a partial fingerprint image lifted from a crime scene by
a forensic expert. Compared to a rolled fingerprint, the latent is
most of the times of bad quality and hard to process
Different fingerprint impressions
 On-line techniques
Online:
• On-line techniques produce the digital image directly from a
subject’s fingertip via digital imaging technology that circumvents
the need for obtaining an impression on a substrate. The resulting
fingerprint image is referred to as a live-scan fingerprint.
• A typical fingerprint scanner comprises:
• a sensor to read the ridge pattern on the finger surface;
• an A/D (Analog to Digital) converter to convert the signal;
• an interface module responsible for communicating with
external devices.
• Almost of all existing sensors belong to one of the following
families: optical, solid-state and ultrasound. These sensors are also
called touch sensors.
Different ways of acquiring
 Different sensors
(a) Optical Frustrated Total Internal Reflection
(FTIR)
(b) Capacitance
(c) Ultrasound Reflection
(d) Piezoelectric Effect
(e) Temperature Differential
 FTIR Based Fingerprint Sensing
• Optical frustrated total internal reflection
• This technique utilizes a glass platen LED light
source and a CCD Camera for constructing
fingerprint images.
• When finger is placed on one side of a glass
platen ( prism) only the ridges of the finger are in
contact with the platen not the valleys.
• The light source illuminates the glass at a certain
angle and the camera is placed such that it can
capture the light reflected from the glass.
FTIR Based Fingerprint Sensing
• The light incident on the ridges is randomly
scattered ( results in dark image) while the
light incident on the valleys suffers total
internal reflection(results in bright image)
• Difficult to have this arrangement in compact
form, since the focal length of small lenses can
be very large.
FTIR BASED SENSING
 CAPCITANCE
• Capacitance based solid state live scan fingerprint
sensors are more commonly used than the
Optical FTIR sensors since they are very small in
size and can be easily embedded into laptop
,mobile phones and computer peripherals.
• Essentially consists of an array of electrodes.
• There are tens of thousands of small capacitance
plates(electrodes ) embedded in chip.
• Fingerprint skin acts as the other electrode,
thereby forming a miniature capacitor
CAPCITANCE
 Ultrasound reflection
• It is based on sending acoustic signals toward the
fingertip and capturing the echo signal.
• The sensors has two main components the
sender that generates short acoustic pulses,and
the receiver that detects the responses obtained
when these pulses bounce off the fingerprint
surface.
• Resilent to dirt and oil accumulations that may
visually mar the fingerprint
• Expensive not suited for large scale production
 Temperature differential
• This mechanism is made up of pyro electric
material that generates a current based on
temperature differentials.
• Temperature differential is created when two
surfaces are brought into contact.
• Fingerprint ridges,being in contact with the
sensor surface, produce an different
temperature differential than valleys that are
away from the sensor surface.
 Fingerprint scanner operation
methods
• Plain fingerprint is obtained by simply placing the
finger on the surface of fingerprint sensor.
• A rolled fingerprint is obtained by rolling the
finger from nail to nail on the surface of
fingerprint sensor
• Sweep fingerprint is obtained by combining
narrow fingerprint slices .( typically 3mm wide)
while the user swipes his finger across the sensor
• Clarity of ridge pattern is another important
determinant of quality.
 Feature extraction
• Commercial fingerprint recognition systems
are mainly based on Level 1 (ridge orientation
and frequency) and Level 2 features (ridges
and minutiae).
• Generally, Level 1 features are first extracted
and then Level 2 features are extracted with
the guidance of Level 1 features.
Feature extraction
 Steps
Typical feature extraction algorithm which
includes four main steps, namely
(a) ridge orientation and frequency estimation
(b) ridge extraction
(c) singularity extraction
(d) Binarization and Thinning
1. Ridge orientation and frequency
Estimation
 2. Singularity Extraction
• Fingerprint singularity can be extracted from
the orientation field using the well known
Poincare index method.
• Poincare index refers to the cumulative
change of orientations along a closed path in
an orientation field.
• To accurately detect the location and type of
singularity, Poincare index is generally
evaluated using the eight neighbors of a pixel.
 Types of Finger Prints
Poincare index of a pixel corresponding to a singular
point can take one of four possible values:
0 (non-singular),
1 (loop),
-1 (delta), and
2 (whorl).
The non-singular, loop, delta and whorl are types of
finer prints
 3. Ridge Extraction

• Since minutiae are special points on ridges, it is natural to

first extract ridges and then detect the minutiae on
ridges.
• Since ridges are darker than valleys, a straightforward
method to detect ridges is to classify any pixel as a ridge
pixel if its gray value is lower than a threshold (for
example, the mean of its local neighborhood).
• However, for most fingerprint images, this method does
not work well for the following reasons: (a) pores on
ridges are brighter than the surrounding pixels; (b) ridges
can be broken due to cuts or creases; (c) adjacent ridges
may appear to be joined due to moist skin or pressure.
 Binarization and Thinning
• The enhanced image can be converted to a
binary image by using either a global threshold
(for example, using the mean pixel value of the
enhanced image) or thresholds that are locally
computed.
• A morphological operation, called thinning, is
used to reduce the ridge width to one pixel.
• Thinning is a common technique in image
processing, which involves iteratively removing
outer ridge pixels
Binarization and thinning
Three steps of matching
Minutiae matching algorithm
 Alignment
• Since two impressions of the same finger taken
at different instances could differ due to
different placement of the finger on the sensor,
an alignment process is required to transform
them to the same coordinate system.
• This process, also known as registration,
transforms one image in such a way that it is
geometrically aligned with the other.
• First, we need to specify a spatial transformation
model
 PAIRING
• After the two minutiae sets are aligned, the
corresponding minutiae are paired.
• A minutia a in the template (reference) minutiae
set is said to be in correspondence with minutia b
in the query minutiae set if and only if their
distance is within a predefined distance
threshold (say, 15 pixels) and the angle between
their directions is within another predefined
angle threshold (say, 20 degrees)
 SCORE
• We need to compute a match score, which is
then compared to a predefined threshold to
classify the two fingerprints as a genuine
match (they come from the same finger) or an
impostor match (they come from two
different fingers).
• This problem can be viewed as a two-class
classification problem with genuine match as
class-1 and impostor match as class-2.
MATCH SCORE
 Palm Print Recognition System
Features of Palm Print
• Friction ridge patterns on the palm of the hand and the
sole of the foot have also been claimed to be unique and
permanent and thus can be used for personal
identification.
• Palm prints and Sole prints have fewer applications than
fingerprints since it is not as convenient to capture them
compared to fingerprints.
• In fact, sole prints are only used to register newborns in
a hospital since it is easier to capture the sole print of
newborns than fingerprints or palmprints (newborns
tend to keep their fists closed!).
 PALMPRINT
• Forensics application is concerned, the main
benefit of using palmprints is in latent
palmprint matching since it is estimated that
about 30% of the latent found at crime scenes
are those of palms.
Palm print features
 Challenges in Palm Print Recognition
• Matching palmprints is more complicated
than matching fingerprints.
• First of all, in terms of image acquisition, it is
more challenging to capture palmprints than
fingerprints since the sensors are larger and
more expensive and greater user cooperation
is required to ensure a good quality palmprint
image due to the concavity of the palm
surface.
Friction ridge patterns
Palm Print Recognition Framework
 Palm Print Recognition
• The palm vein images employed in identification using
palm vein obtained under near-infrared illumination
(NIR) and the images normally emerge darker with low
contrast.
• So, image development more evidently demonstrates
the vein and texture patterns.
• First, approximate the condition intensity profiles by
segmenting the image into a little overlapping. In 32 X
32 blocks, three pixels are overlapped between two
blocks to address the blocky effect.
• The average gray-level pixels in each block are
forecasted.
 Palm Print Recognition
The main modules of a palm-print verification
system preforms
a) Palm-print senses the palm print of a human
obtained by a palm print scanner.
b) Pre-processing in which the input palm print is
improved and adapted to shorten the process of
feature extraction;
c) Feature extraction in which the palm print is
further processed to produce discriminative
features also called feature vectors
d) Matching in which the feature vector of the
input palm print compared against with one or
more existing templates.
 Face recognition

Face recognition and acquisition

Face detection, feature extraction
and matching
 Face recognition
• Face recognition can be defined as the process of
establishing a person’s identity based on their
facial characteristics
• The problem of face recognition involves
comparing two face images and determining if
they are of the same person
• Face images of a person may have variations in
age, pose, illumination, and facial expressions
exhibit changes in appearance due to make-up,
facial hair, or accessories (e.g., sunglasses, hat
etc)
 ADVANTAGES
• Face can be captured at a longer standoff
distance using non-contact sensors.
• The face conveys not only the identity, but also
the emotions of a person (e.g., happiness or
anger) as well as biographic information(e.g.,
gender, ethnicity, and age).
• Face recognition has a wide range of applications
in law enforcement, civilian identification,
surveillance systems, and entertainment/
amusement systems
 Level 1 details
• Consists of gross facial characteristics that are easily
observable.
• Examples include the general geometry of the face and
global skin color. Such Features can be used to quickly
discriminate between
• (a) a short round face and an elongated thin face;
• (b) faces exhibiting predominantly male and female
characteristics;
• (c) faces from different races.
• These features can be extracted even from low
resolution face images (< 30 interpupilary distance
(IPD)1).
 Level 2
• Level 2 details consist of localized face
information such as the structure of the face
components (e.g., eyes, nose, mouth etc), the
relationship between facial components and the
precise shape of the face.
• These features are essential for accurate face
recognition, and they require a higher resolution
face image (30 to 75 IPD).
• The characteristics of local regions of the face can
be represented using geometric or texture
descriptors.
 Level 3
• Level 3 details consist of unstructured, micro
level features on the face, which includes
scars, freckles, skin discoloration, and moles.
• One challenging face recognition problem
where Level 3 details may be critical is the
discrimination of identical twins
All three levels of facial features
Design of a face recognition system
A typical face recognition system is composed of
three modules:
(a) image acquisition,
(b) face detection, and
(c) face matching
 Image acquisition
Face image acquired from a sensor can be
categorized based on
a) the spectral band (e.g., visible, infrared, and
thermal) used to record the image
b) the nature of the image rendering technique
(e.g., 2D, 3D, and video).
• Since most of the automated face recognition
systems make use of 2D images acquired in
the visible spectrum
 Face matching
• Face matching is usually carried out by
comparing the features extracted from the
probe and gallery images
 FACE DETECTION
• Face detection is the first step in most face-
related applications including face recognition,
facial expression analysis, gender/ethnicity/age
classification, and face modeling.
• There are a number of approaches for detecting
faces in a given image
• State-of-the-art face detection methods are
typically based on extracting local texture
features from the given image and applying a
binary (two-class) classifier to distinguish
between a face and non-face
Face Detection:
Viola-Jones face detector: The Viola-Jones face
detector scans through the input image with detection
windows of different sizes and decides whether each
window contains a face or not.
• Viola-Jones face detector is not perfect and can
produce both false positive and false negative errors
• A false positive error refers to the detection of a
face where none exists, while a false negative error
indicates that a face present in the image was not
detected.
Viola-Jones face detector
 Viola Jones Face Detector
• The Viola-Jones face detector scans through the
input image with detection windows of different
sizes and decides whether each window contains
a face or not.
• Figure shows the scanning process ranging from
small to large windows.
• In each window, the existence of a face candidate
is decided by applying a classifier to simple local
features derived using rectangular filters
 Feature Extraction and Matching

There are three main approaches to match the

detected face images.
1. Appearance-based,
2. Model-based, and
3. Texture-based methods
 Appearance based
• Appearance-based techniques generate a compact
representation of the entire face region in the acquired
image by mapping the high-dimensional face image
into a lower dimensional sub-space
• This sub-space is defined by a set of representative
basis vectors, which are learned using a training set of
images the mapping can be either linear or non-linear,
commonly used schemes such as
• Principal Component Analysis (PCA),
• Linear Discriminant Analysis (LDA), and
• Independent Component Analysis (ICA)
• involve linear projections
 Model based techniques
• Model-based techniques attempt to build 2D
or 3D face models that facilitate matching of
face images in the presence of pose variations.
• While the Face Bunch Graphs (FBG) and Active
Appearance Model (AAM) are examples of 2D
face models, the morphable model is a 3D
model.
 Texture-based methods

• Texture-based approaches try to find robust

local features that are invariant to pose or
lighting variations.
• Examples of such features include gradient
orientations and Local Binary Patterns (LBP).
CATEGORIZATION OF FACE RECOGNITION TECHNIQUES
Appearance-based face recognition

• Principal Component Analysis

• Linear Discriminant Analysis

• Principal Component Analysis:

• Principal Component Analysis (PCA) is one of the
earliest automated methods proposed for face
recognition. PCA uses the training data to learn a
subspace that accounts for as much variability in the
training data as possible. This is achieved by
performing an Eigen value decomposition of the
covariance matrix of the data.
• PCA can be referred to as an unsupervised
learning method, because the class label (user
identity information) is never used during the
learning of the basis faces.
• Hence, the face recognition accuracy based on
PCA cannot be expected to be very High
 Linear Discriminant Analysis
• Linear Discriminant Analysis (LDA) explicitly uses
the class label of the training data and conducts
subspace analysis with the objective of
minimizing intra-class variations and maximizing
inter-class variations
LDA
• can generally be expected to provide more
accurate face recognition when sufficient face
image samples for each user are available during
training.
 Model-based face recognition
• Model-based techniques try to derive a pose-
independent representation of the face
images that can enable matching of face
image across different poses.
• These schemes typically require the detection
of several fiducial or landmark points in the
face (e.g., corners of eyes, tip of the nose,
corners of the mouth, homogeneous regions
of the face, and the chin)
Model-based face recognition
Elastic Bunch Graph Matching: The Elastic Bunch Graph
Matching (EBGM) scheme represents a face as a labeled graph
with
> Each node of the graph is labeled with a set of Gabor
coefficients (also called a jet) that characterizes the local texture
information around the landmark point, and
> The edges of the graph is labeled based on the distance
between the corresponding fiducial points
> The Gabor coefficient at a location in the image can be
obtained by convolving the image with a complex 2D Gabor filter
centered at that location.
> By varying the orientation and frequency of the Gabor filter, a
set of coefficients or a Gabor jet can be obtained
A Face Bunch Graph (FBG) model can be constructed
in two stages from a training set of face images with a
specific pose

• In the first stage, the designer has to manually

mark the desired fiducial points and define the
geometric structure of the image graph for one (or
a few) initial image(s).
• The image graphs for the remaining images in the
training set can be obtained semi-automatically, by
comparing the new images to model graphs
(images that have been already marked) based on
the extracted Gabor jets.
• During this process, manual intervention is
required only if the fiducial points are identified
incorrectly
• Since all the training images have the same pose,
the graphs corresponding to these face images will
have the same structure
In the second stage, a FBG is obtained from the
individual image graphs by combining a representative
set of individual graphs in a stack-like structure.
• Each node in the FBG is labeled by a set of Gabor
jets that represent the local variations in the
associated fiducial point among the population of
users in the training set
• For example, an eye bunch may include jets from
open, closed, male and female eyes, etc. that
cover the variations in the local structure of the
eye.
• An edge between two nodes of the FBG is labeled
based on the average distance between the
corresponding nodes in the training set.
• Typically, a separate FBG is constructed for each
pose and the correspondence between the nodes
of bunch graphs belonging to different poses is
specified manually.
 STEPS
1.Find the approximate face position by coarsely
scanning the input image with a condensed FGB
(an average graph obtained by taking the mean
Gabor jets at each bunch).
• This is achieved by extracting the Gabor jets at
some discrete locations in the given image
and comparing it to the jets in the condensed
FBG, while taking into account the geometric
structure of the FBG
 STEPS
2. Refine the position and size of the face by
searching the image again with the full FBG, whose
size and aspect ratio is systematically varied.
• When computing the similarity between a Gabor
jet in the given image and a bunch of jets in the
FBG, only the FBG jet that best matches with the
given image jet is considered.
3. Precisely locate the fiducial points by moving all
the nodes locally and relative to each other to
optimize the graph similarity further.
FBG
Texture-based face recognition
• Texture of an image uses the distribution of local pixel values.
• Scale Invariant Feature Transformation (SIFT) and Local
Binary Pattern (LBP) are two most well known schemes for
analysis of local texture
Scale Invariant Feature Transform: Scale Invariant Feature
Transform (SIFT) is one of the most popular local representation
schemes used in object recognition.
(a) key point extraction,
(b) descriptor calculation in a local neighborhood at each key
point.
Texture-based face recognition

• Just like the fiducial points in the model-based

approach, the key points can be used to achieve
tolerance against pose variations
• The face image is typically divided with multiple
patches and the SIFT descriptor is constructed from
each patch.
• The final descriptor is obtained by concatenating all
the descriptors from all the patches
 LBP
Local Binary Pattern
• Local Binary Pattern (LBP) has been successfully
used as a local texture descriptor in general object
recognition as well as in face recognition
• LBP features are usually obtained from image pixels
of a 3×3 neighborhood region
• The basic LBP operator compares the 8 neighboring
pixel intensity values to the intensity value of the
central pixel in the region and represents the result
as a 8-bit binary string.
 MLBP
Multi-scale LBP (MLBP) is an extension of the basic LBP
• MLBP introduces a radius parameter R, which means that the
compared neighbors are R pixels away from the center pixel.
There is also another parameter P, which is the number of
sampling points along the circle of radius R.
• If a sampling point is off the pixel grid, bilinear interpolation
of pixel values can be applied to obtain the intensity value of
the sampling point.
The MLBP operator with parameter
• Generally,MLBP with larger value of P provides more detailed
information about the local region. However, when P
becomes larger, the dimension of the descriptor also
increases.
• MLBP operators with different values of R encode different
local image structures, ranging from micro to macro details
 IRIS
• Iris recognition and acquisition
• Iris Segmentation, normalization and
matching
 IRIS RECOGNITION
• The use of the ocular region as a biometric
trait has gained importance, especially due to
significant advancements made in iris
recognition since 1993.
• The ocular region of the human face consists
of the eyes and the surrounding structures
such as facial skin, eyebrows, and nose bridge
 IRIS
• The iris is an internal organ of the eye that is
located just behind the cornea and in front of
the lens.
• The primary function of the iris is to regulate
the amount of light entering the eye by
dilating or contracting a small opening in it
called the pupil.
• The iris contracts the pupil when the ambient
illumination is high and dilates it when the
illumination is low.
IRIS
 LAYERS OF IRIS
• The iris is a multilayered structure and a cross-
section of the iris reveals the following layers:
• The posterior layer: at the back, which is two
cells thick, contains heavily pigmented
epithelial cells, making it impenetrable to
light.
• The muscle layer: Above it consists of the
sphincter and dilator muscles that contract
and dilate the pupil
 LAYERS OF IRIS
• The Stromal layer: located above the muscles,
is made up of collagenous connective tissue
(arranged in an arch-like configuration) and
blood vessels (arranged along the radial
direction).
• The anterior border layer: is the foremost
layer and has an increased density of
chromatophores (i.e., pigment containing
cells) compared to the stromal layer
 ANTERIOR PORTION
• The anterior portion of the iris - consisting
collectively of the muscles, stroma, and the
border layers - is the foremost visible portion
of the iris.
• Therefore, it can imaged by a camera and is
the focus of all automated iris recognition
systems.
 IRIS IMAGE
• Figure shows an image of the iris as viewed by
a near-infrared camera.
• The iris, from this perspective, appears as an
annular entity bound by the pupillary
boundary (that separates it from the dark
pupil) and the limbus boundary (that
separates it from the white sclera).
IRIS IMAGE
 IRIS ZONES
• The iris image is partitioned into two zones: the central
pupillary zone and the surrounding ciliary zone.
• These two zones are divided by a circular zigzag ridgeline
known as the collarette.
• Many pit-like irregular structures appear mainly in the
region around the collarette. These structures are called
crypts (Fuchs crypts) and they permit fluids to quickly
enter and exit the iris during dilation and contraction of
the pupil.
• Near the outer part of the ciliary zone, concentric lines
can be seen, especially in case of darkly pigmented
irides1. These lines become deeper as the pupil dilates
and are called contraction furrows. In the pupillary zone,
radial furrows are observed.
IRIS ZONES
 IRIS
• The color of the iris is primarily defined by the
pigmentation present in it. The pigmentation itself
is controlled by the number of melanin granules - a
genetically determined factor.
• However, other factors, such as the cellular density
of the stroma, can also affect the color of the iris.
• The color of the iris does not play a significant role
in iris recognition systems.
• It is the texture detail present in the anterior
portion of the iris that is useful for recognition.
 Design of an Iris Recognition System
• An iris recognition system may be viewed as a pattern
matching system whose goal is to compare two irides
and generate a match score indicating their degree of
similarity or dissimilarity.
• Thus, a typical iris recognition system has four different
modules:
1. The acquisition
2. Segmentation
3. Normalization, and
4. Encoding/matching modules.
BLOCK DIAGRAM OF IRIS RECOGNITION SYSTEM
 ACQUISITION
• Acquisition: The role of the acquisition module is
to obtain a 2D image of the eye using a mono
chrome CCD camera that is sensitive to the near-
infrared (NIR) range of the electromagnetic
spectrum.
• An external source of NIR light, often co-located
with the acquisition system, is used to illuminate
the iris.
• Most iris recognition systems require the
participating individual to be cooperative and to
place their eye in close proximity to the camera.
 SEGMENTATION
• Segmentation: The segmentation module
localizes the spatial extent of the iris in the image
of the eye by isolating it from other structures
present in its vicinity. These structures include the
sclera, the pupil, the eyelids, and the eyelashes.
• Typically, segmentation is accomplished by
detecting the inner and outer boundaries of the
iris (commonly referred to as the pupillary
boundary and the limbus boundary, respectively),
and the eyelids and eyelashes that can interrupt
the circular contour of the limbus boundary.
 SEGMENTATION
• These segmentation techniques model iris
boundaries and the two eyelids with simple
geometric models.
• Pupil and limbus are often modelled as circles
and the two eyelids are modelled as parabolic
arcs.
• It is desired to distinguish the iris texture from
the rest of the image.
• The integro-differential operator, Hough
transform and active contour models are
commonly used to detect the boundaries of the
iris.
SEGMENTATION
 NORMALIZATION
• Normalization: Once the inner and outer
boundaries of the iris are estimated, a geometric
normalization scheme is invoked to transform the
iris texture within the annular region from
cartesian coordinates to pseudo polar
coordinates via a Rubber Sheet Model. This
process is called as the “unwrapping of the iris”
• It results in a rectangular entity whose rows
correspond to the angular direction in the
original iris and whose columns correspond to its
radial direction.
NORMALIZATION
 Encoding and Matching:
• Encoding and Matching: While the unwrapped iris may be directly
used to compare two irides (e.g., by using correlation filters),
typically a feature extraction routine is used to encode its textural
content.
• Most encoding algorithms perform a multi-resolution analysis of
the iris by applying wavelet filters and examining the ensuing
response.
• A commonly used encoding mechanism uses quadrature 2D
Gabor Wavelets to extract the local phasor information of the iris
texture.
• Each phasor response is then encoded using two bits of information
based on the quadrant of the complex plane in which it lies. The
resulting 2D binary code is referred to as the iris code
• Two such iris codes can be compared using the
Hamming distance, which computes the
number of corresponding bits that are different
across them;
• The binary mask computed in the segmentation
module is used to ensure that only bits
corresponding to valid iris pixels are compared.
 Ear
• Ear recognition
• Ear detection
 Ear
• The human ear is observed to exhibit
variations across individuals as assessed by
the curves, surfaces, and geometric
measurements pertaining to the visible
portion of the ear, commonly referred to as
the pinna.
• The structure of the pinna depicting various
anatomical features can be seen in Figure
EAR
 The Ear offers several advantages
(a) the structure of the ear has been observed
to be stable despite aging, and ear growth is
almost linear after the age of four
(b) the ear, unlike other facial features, is
minimally impacted by changes in facial
expression; and
(c) image acquisition does not involve explicit
contact with the sensor.
 EAR RECOGNITION SYSTEM
A typical ear recognition system consists of the following
components:
(a) An ear detection module (also known as segmentation)
that localizes the position and spatial extent of the ear in an
image
(b) a feature extraction module that extracts discriminative
features from the ear
(c) a matching module that compares the features extracted
from two ear images and generates a match score; and
(d) a decision module that processes the match score(s) and
establishes the identity of the subject.
 Ear detection
• 1. Template Matching
• 2. Model-based Detection
• 3. Morphological-operator-based Detection
• 4. Face-geometry-based Detection
 TEMPLATE MATCHING
Template Matching:
• In a template matching scheme, a template of a
typical ear is constructed and is matched with
each location in the query image.
• The location giving the highest score is
considered as the region containing the ear.
• The template may consist of an edge image of
the ear or a set of descriptors extracted from the
ear such as the response to a set of filters or a
histogram of shape curvatures in case a 3D
image of the ear is being used for recognition
 Model based detection
• A model-based detection technique assumes
certain characteristics of the shape of the ear
and tries to find regions that manifest such
characteristics.
• The shape of the helix, for example, is usually
elliptical so a generalized Hough transform
tuned for detecting ellipses can be used to
locate the ear in an edge image.
 Morphological-operator-based
Detection
• Since the structure of the ear is usually more
intricate than the structure of the remaining
region in a profile face image, morphological
transformations such as the Top-hat
transformation can be used.
• A Top-hat transformation essentially subtracts
a morphologically smoothened version of an
image from itself, thereby highlighting finer
details.
 Face-geometry-based Detection

• Since in a profile image the nose can be easily

detected as the point with high curvature, it is
possible to constrain the search for the ear in
an appropriate location relative to the nose.
 Ear recognition
• Subspace analysis-based techniques
• Sparse representation-based techniques
• Point set matching-based techniques
• Image filtering-based techniques
• Geometric measurements-based techniques
• Transformation-based techniques
• 3D techniques
Subspace analysis-based techniques
• Projecting the ear image onto a set of principal
directions is an effective way to obtain a salient
and compact representation of an ear.
• Subspace projection techniques such as PCA, and
LDA have been successfully used in literature for
matching ear images.
• Manifold learning based techniques such as
Locally Linear Embedding (LLE) and Kernel PCA
have also been used to perform ear recognition.
Sparse representation-based techniques

• Optimization techniques that minimize L-1 norm

of the distance vector between the transformed
query and all the transformed templates in a
database have been shown to provide high
recognition accuracy in object recognition
studies.
• This technique has also been successfully used
for ear recognition.
 Point set matching-based techniques
• Elastic bunch graph matching is an effective technique
to recognize faces based on responses to a bank of
Gabor filters at some nodal points on the face.
• This technique has also been successfully used for
matching ear images where a number of landmark
points can be easily detected, thanks to its complex
structure.
• Scale Invariant Feature Transform (SIFT) is a well
known technique for matching two images where a set
of salient points can be reliably and repeatably
extracted from them.
 Image filtering-based techniques
• In certain techniques, the ear image is first
enhanced to highlight the discriminative
features and suppress the noise.
• Two common techniques that use this basic
procedure are
a) force field transformation and
b) local binary patterns
 Force field transformation
• A force field transformation obtains the intensity of
the forces at each location in the image where each
pixel is considered as a source of the force with
intensity proportional to its value.
• Force field transformation has been shown to
effectively remove the noise in the ear image leading
to significant improvement in recognition accuracy.
• Further, a set of lines indicating the gradients of this
force field can be extracted and used for matching.
• Figure for a depiction of force field extracted from an
ear image with the force field lines marked.
 Local binary patterns
• Local binary patterns essentially characterize each
pixel based on variation of intensity of that pixel
along a set of directions.
• The variation along each direction is encoded as a
single bit indicating whether the intensity is
increasing or decreasing, and the set of bits
associated with each direction is used to obtain an
integer value for each pixel.
• Such a transformation effectively reduces the effect
of illumination variations and other sources of
noise, thereby generating an enhanced image which
can be used for robust matching.
 Geometric measurements-based techniques
• Features obtained by measuring certain
geometric characteristics of the ear can also be
used as a set of discriminative features.
• As an example, the centroid of a near image
obtained from its edge image can be used as a
center to draw concentric circles with pre-
specified radii.
• Various measurements, such as number of
points on a circle intersecting the edge image
or distance between two consecutive
intersections, can be used as a feature vector.
 Transformation-based techniques
• Various image transformation techniques such
as Fourier transform or wavelet transform
can also be applied to extract discriminative
features from an ear image.
• Fourier transform can also be applied in order
to obtain a rotation and translation invariant
representation of the ear –
• For example, by using a polar coordinate
system and extracting only the magnitude of
the Fourier transform.
 3D Techniques
• 3D images offer depth information that can be
used in conjunction with the 2D texture information
to improve the recognition accuracy.
• In the case of 3D ear images, local histograms of
shape curvature values can be used to match two
ear images.
• The Iterative Closest Point (ICP) algorithm is
commonly used to register and match 3D ears.
• In the ICP technique, each point in the input ear
image is used to obtain a corresponding point in the
template image and the input is then rotated and
translated in order to minimize the distances
between corresponding points.
 Challenges in ear recognition
• There are no commercial biometric systems at
this time that explicitly utilize features of the ear
for human recognition.
• But the performances of ear recognition
algorithms have been tested on some standard
ear datasets. Experiments suggest that ear
images obtained under controlled conditions can
result in good recognition accuracy.
• 1. Ear Occlusion
• 2. Earprint Identification
 Ear Occlusion
• One of the main challenges faced by an ear
recognition system is occlusion due to the
subject’s hair.
• One way to address such occlusion is by capturing
the thermogram along with the visible light
image.
• In a thermogram, the hair can be easily detected
(and possibly isolated) as its temperature is
usually lower than that of the skin.
 Ear print identification
• Earprints, or earmarks, are marks left by
secretions as a result of pressing the ear against a
flat surface.
• These marks mainly consist of impressions of the
helix, anti-helix, tragus, and anti-tragus.
• Other features include the earlobe and the crus
of helix, but they are less frequently observed.
• Earprints can be compared based on details such
as the notches and angles in imprinted samples,
the positions of moles, folds, and wrinkles, and
the position of pressure points
• Earprint identification is usually manually
performed by identifying and matching a set
of geometric features from the earprint such
as the intersection points of ear curves with a
regular grid or locations of certain other
landmark points.
 Hand Geometry
• Hand geometry, as the name suggests, refers to the geometric
structure of the hand.
• This structure includes width of the fingers at various locations,
width of the palm, thickness of the palm, length of the fingers,
contour of the palm, etc.
• Although these metrics do not vary significantly across the
population, they can still be used to verify the identity of an
individual.
• Hand geometry measurement is non-intrusive and the verification
involves a simple processing of the resulting features.
• Unlike palmprint, this method does not involve extraction of
detailed features of the hand
 Hand Geometry
A typical hand geometry system consists of four main
components:
• Image acquisition,
• Hand segmentation and alignment,
• Feature extraction,
• Feature matching.
 Image capture
• Contact-based vs Contactless:
– A typical system requires the user to place her
hand on a flat surface prior to capturing. Such
systems are contact-based and require explicit
cooperation of the subject being identified.
– To address these issues, contactless recognition
systems have been proposed. However, such
systems are required to address the intra-class
variability in the captured images due to the
articulation of the hand in all three dimensions.
 Image capture
• Dorsal vs Palmar
– An image of the hand is acquired by placing the
hand on a flat surface and imaging the back of
the hand with a CCD camera.
– However, there has been interest in capturing the
ridge patterns present on the palm and fingers
along with the hand shape by imaging the
palmar aspect of the hand
– Drawback of such a system is that it is rather
inconvenient for a user to place the hand on a
platen with the palm facing upward.
 Image capture
• Peg-based vs Pegless:
– In order to guide the positioning of the hand on
the platen for imaging purposes (e.g., to prevent
the fingers from touching each other), a few
pegs are usually placed on the sensor platen.
– It adds to the complexity of using the system
and, thus, greater inconvenience to users.
 Image capture
• The image acquisition system comprises of a
light source, a camera, a single mirror and a
surface (with five pegs on it)
• Pegs serve as control points for appropriate
placement of the right hand of the user
• A 640 by 480 8-bit grayscale image of the
hand is captured
 Hand segmentation
• The hand boundary must be extracted in order to
determine the region of interest.
• The segmented hand may be further divided to
obtain smaller segments corresponding to
individual fingers.
• Features can then be extracted separately from
each of these segments.
 Feature Extraction
• Two kinds of features are
extracted from a hand or a
finger silhouette:
– one-dimensional
geometric measurements
and two-dimensional
shape-based features.
• The geometric
measurements include
length and width of
fingers, length and width
of the palm, and thickness
of the fingers.
 Feature Extraction
• It is also possible to use the set of
points along the contours of the
silhouette (or the segmented
image itself) as features.
• The dimensionality of these
features can be reduced in order
to obtain a more discriminative
and compact representation of the
hand.
• 16 axes along which the various
features mentioned above have
been measured
• The hand is represented as a
vector of the measurements
selected above F = ( f1, f 2 ,..., f16 )
 Feature Extraction
• X = (x1, x2 , x3 ,..., xLen ) is the gray values of the pixels along
a axis
• A sequence of pixels along a measurement
axis will not have an ideal gray scale profile
– Background lighting, color of the skin, noise and
etc.
 Feature Extraction
• Result
 Feature matching
• The features extracted from a segmented hand
image can often be denoted as a feature vector
in the Euclidean space.
• Common distance measures such as Euclidean
and Manhattan distances can be effectively
used to compare two hand images
• Mahalanobis distance
• ML algorithms such as SVM
 Why Hand Geometry
• What is the most effective biometric measurement?
– Each biometrics has its strengths and limitation
– It’s depend on application
– User acceptability is the most significant factor
• For many access control applications, like dormitory
meal plan access, very distinctive biometrics, e.g.,
fingerprint and iris, may not be acceptable
• Hand geometry-based verification systems are not new
and have been available since the early 1970s
 A Biometric Identification System Based on
Eigen-palm and Eigen-finger Features
• The identification process can be divided into
the following phases:
– Capturing the image
– Preprocessing, extracting and normalizing the
palm and strip-like finger subimages
– Extracting the eigenpalm and eigenfinger features
based on the PCA
– Matching and fusion
– A decision based on the (k, l)-NN classifier and
thresholding
System Description
Preprocessing and Sub-images Extraction

(a) Example of a scanned hand image. (b) Binerized

hand image.
 Preprocessing and Sub-images
Extraction

(a)The hand contour and the relevant points for finding the
regions of interest. (b) Processed finger on the hand contour.
 Geometry and Lighting Normalization

(a) Original hand image with the regions of interest marked on it. (b)
Palm subimage. (c) Little-finger subimage. (d) Ring-finger subimage.
(e) Middle-finger subimage. (f) Index-finger subimage. (g) Thumb
subimage.
 Geometry and Lighting Normalization
• After the subimages have been extracted, a lighting
normalization using histogram fitting is applied.
• In this process, a target histogram G(l) is selected for each of
the six subimage

Subimages (a) before and (b) after the histogram fitting.

 Lighting Normalization

Palm sub-images with the corresponding histograms (a)

before and (b) after the histogram fitting.
Extraction of the Eigenpalm and Eigenfinger Features

Average palm and finger

images from database.
Eigenpalms and eigenfingers obtained from the
database (indicated by the corresponding ordinal
numbers): (a) the eigenpalms and (b) the little-
finger eigenfingers.
 MATCHING AND FUSION AT THE
MATCHING-SCORE LEVEL
• The matching between corresponding feature vectors is based
on the Euclidean distance
• The distances are normalized and transformed into similarity
measures
– The frequency distribution of the distances of templates of the same
person is calculated.