33

CHAPTER 2

REVIEW OF LITERATURE

Research on biometric methods has gained renewed attention in recent

years brought on by an increase in security concerns. The recent world attitude

towards terrorism has influenced people and their governments to take action

and be more proactive in security issues. This need for security also extends to

the need for individuals to protect, among other things, their working

environments, homes, personal possessions and assets. Many biometric

techniques have been developed and are being improved with the most

successful being applied in everyday law enforcement and security

applications. Biometric methods include several state-of-the-art techniques.

Among them, fingerprint recognition is considered to be the most powerful

technique for utmost security authentication.

Advances in sensor technology and an increasing demand for biometrics

are driving a burgeoning biometric industry to develop new technologies. As

commercial incentives increase, many new technologies for person

identification are being developed, each with its own strengths and weaknesses

and a potential niche market. This chapter reviews some well-known

biometrics with special emphasis to fingerprint.

2.1. BIOMETRIC SECURITY : FROM 1950-2012

The term “Biometrics” is derived from the Greek words “bio” (life) and

“metrics” (to measure) (Rood and Hornak, 2008). Automated biometric

systems have only become available over the last few decades, due to the

significant advances in the field of computer and image processing. Although

biometric technology seems to belong in the twenty first century, the history of

biometrics goes back thousands of years. The ancient Egyptians and the

Chinese played a large role in biometrics history. Today, the focus is on using

biometric face recognition, iris recognition, retina recognition and identifying

34

characteristics to stop terrorism and improve security measures. This section

provides a brief history on biometric security and fingerprint recognition.

During 1858, the first recorded systematic capture of hand and finger

images for identification purposes was used by Sir William Herschel, Civil

Service of India, who recorded a handprint on the back of a contract for each

worker to distinguish employees (Komarinski, 2004).

During 1870, Alphonse Bertillon developed a method of identifying

individuals based on detailed records of their body measurements, physical

descriptions and photographs. This method was termed as “Bertillonage” or

anthropometrics and the usage was aborted in 1903 when it was discovered that

some people share same measurements and physical characteristics (State

University of New York at Canton, 2003).

Sir Francis Galton, in 1892, developed a classification system for

fingerprints using minutiae characteristics that is being used by researchers and

educationalists even today. Sir Edward Henry, during 1896, paved way to the

success of fingerprint recognition by using Galton's theory to identify prisoners

by their fingerprint impressions. He devised a classification system that

allowed thousands of fingerprints to be easily filed, searched and traced. He

helped in the first establishment of fingerprint bureau in the same year and his

method gained worldwide acceptance for identifying criminals (Scottish

Criminal Record Office, 2002).

The concept of using iris pattern for identification was first proposed by

Ophthalmologist Frank Burch in 1936 (Iradian Technologies, 2003). During

1960, the first semi-automatic face recognition system was developed by

Woodrow W. Bledsoe, which used the location of eyes, ears, nose and mouth

on the photographs for recognition purposes. In the same year, the first model

of acoustic speech production was creased by a Swedish Professor, Gunnar

Fant. His invention is used in today's speaker recognition system (Woodward

et al, 2003).

35

The first automated signature recognition system was developed by

North American Aviation during 1965 (Mauceri, 1965). This technique was

later, in 1969, used by Federal Bureau of Investigation (FBI) in their

investigations to reduce man hours invested in the analysis of signatures. The

year 1970 introduced face recognition towards authentication. Goldstein et al.

(1971) used 21 specific markers such as hair color, lip thickness to automate

the recognition process. The main disadvantage of such a system was that all

these features were manually identified and computed.

During the same period, Dr.Joseph Perkell produced the first behavioral

components of speech to identify a person (Woodward et al, 2003). The first

commercial hand geometry system was made available in 1974 for physical

access control, time and attendance and personal identification. The success of

this first biometric automated system motivated several funding agencies like

FBI Fund, NIST for the development of scanners and feature extraction

technology (Ratha and Bolle, 2004), which will finally lead to the development

of a perfect human recognizer. This resulted in the first prototype of speaker

recognition system in 1976, which was developed by Texas instruments and

was tested by US Air Force and the MITRE Corporation. In 1996, the hand

geometry was implemented successfully at the Olympic Games and the system

implemented was able to handle the enrollment of over 65,000 people.

Drs. Leonard Flom and Aran Safir, in 1985, found out that no two irises

are alike and their findings were awarded a patent during 1986. In the year

1988, the first semi-automated facial recognition system was deployed by

Lakewood Division of Los Angeles Country Sheriff's Department for

identifying suspects (Angela, 2009). This was followed by several land marked

contributiona by Sirovich and Kirby (1989), Turk and Pentland (1991), Philipis

et al. (2000) in the field of face recognition.

The next stage in fingerprint automation occurred at the end of 1994

with the Integrated Automated Fingerprint Identification System (IAFIS)

36

competition. The competition identified and investigated three major

challenges:

(1) digital fingerprint acquisition

(2) local ridge characteristic extraction and

(3) ridge characteristic pattern matching (David et al., 2005).

The first Automated Fingerprint Identification System (AFIS) was developed

by Palm System in 1993. During 1995, the iris biometric was officially released

as a commercial authentication tool by Defense Nuclear Agency and Iriscan.

The year 2000 envisaged the first face recognition vendor test (FRVT,

2000) sponsored by the US Government agencies and the same year paved way

for the first research paper on the use of vascular patterns for recognition (Im

et al., 2001). During 2003, ICAO (International civil Aviation Organization)

adopted blueprints for the integration of biometric identification information

into passports and other Machine Readable Travel Documents (MRTDs).

Facial recognition was selected as the globally interoperable biometric for

machine-assisted identity confirmation with MRTDs.

The first statewide automated palm print database was deployed by the

US in 2004. The Face Recognition Grand Challenge (FRGC) began in the same

year to improve the identification problem. In 2005, Iris on the move was

announced by Biometric Consortium Conference for enabling the collection of

iris images from individuals walking through a portal.

2.2. FINGERPRINT RECOGNITION SYSTEM – PAST TO PRESENT

Fingerprint imaging technology has been in existence for centuries.

Archaeologists have uncovered evidence suggesting that interest in fingerprints

dates to prehistory.

In Nova Scotia petroglyphs (from the time of pre-historic Native

Americans) showing a hand with exaggerated ridge patterns has been

37

discovered. In ancient Babylon and China, fingerprints were impressed on clay

tablets and seals. The use of fingerprints as a unique human identifier dates

back to second century B.C. China, where the identity of the sender of an

important document could be verified by his fingerprint impression in the wax

seal (Ruggles, 1996). In fourteenth-century Persia fingerprints were impressed

on various official papers. At that time, a governmental official observed that

no two fingerprints were exactly alike.

Using the newly invented microscope, Professor Marcello Malpighi at

the University of Bologna noted ridges on the surface of fingers in 1686. He

described them as loops and spirals but did not note their value as a means of

personal identification. Later, in 1823 at the University of Breslau, Professor

John Evangelist Purkinje published his thesis proposing a system of

classification based on 9 different fingerprint patterns. This was the first step

towards the modern study of fingerprints.

The first modern use of fingerprints occurred in 1856 when Sir William

Herschel, the Chief Magistrate of the Hooghly district in Jungipoor, India, had

a local businessman, Rajyadhar Konai, impress his handprint on the back of a

contract. Later, the right index and middle fingers were printed next to the

signature on all contracts made with the locals. The purpose was to frighten the

signer of repudiating the contract because the locals believed that personal

contact with the document made it more binding. As his fingerprint collection

grew, Sir Herschel began to realize that fingerprints could prove or disprove

identity. Despite his lack of scientific knowledge in fingerprinting he was

convinced that fingerprints are unique and permanent throughout life.

The French anthropologist, Alphonse Bertillon, devised the first widely

accepted scientific method of biometric identification in 1870. The Bertillon

system, Bertillonage, or anthropometry was not based on fingerprinting but

relied on a systematic combination of physical measurements. These, among

others, included measurements of the skull width, foot length, and the length of

38

the left middle finger combined with hair color, eye color, as well as face and

profile pictures. By grouping the data any single person could be placed into

one of 243 distinct categories. For the next thirty years, Bertillonage was the

primary method of biometric identification.

Dr. Henry Faulds, British Surgeon-Superintendent of the Tsukiji

Hospital in Tokyo, took up the study of fingerprints in the 1870’s after noticing

finger imprints on prehistoric pottery. In 1880, in the October 28 issue of the

British scientific periodical Nature, Dr. Faulds was the first to publish a

scientific account of the use of fingerprint as a means of identification. In

addition to recognizing the importance of fingerprints, for identification he

devised a method of classification as well. Dr. Faulds is credited for the first

fingerprint identification-based on a fingerprint left on an alcohol bottle. The

method of classification proposed by Dr. Faulds is called Henry Classification

system and is based on patterns such as loops and whorls, which is still used

today to organize fingerprint card files.

Continuing the work of Dr. Faulds, Sir William Herschel and Sir Francis

Galton established the individuality and permanence of fingerprints. This book,

“Fingerprints” from 1892, contains the first fingerprint classification system

containing three basic pattern types: loop, arch, and whorl. The system was

based on the distribution of the pattern types on the ten fingers, e.g. LLAWL

LWWLL. The system worked, but was yet to be improved with a classification

that was easier to administer. Sir Galton identified the characteristics used for

personal identification, the unique ridge characteristics known as minutiae,

which are often referred to as “Galton’s details”.

In 1892, Juan Vecetich, an Argentine Police official, made the first

criminal fingerprint identification. He was able to identify a woman, who had

murdered her two sons and cut her own throat in an attempt to avoid blame.

Her bloody print was left on a doorpost, proving her identity as the murderer.

39

2.2.1. Shift from Bertillonage to fingerprinting

During the 1890’s, Sir Edward Richard Henry, a British official in

Bengal believed that a fingerprinting system was the solution to his problem of

verifying the identity of criminals. He studied the works of Sir Galton and Sir

Henry and proved that they could be used to produce 1,024 primary

classifications, which was instituted in Bengal in 1897. The system is described

in his book, “Classification and Uses of Finger Prints”. In June 1897,

Bertillonage was replaced and the Henry Classification System became the

official method of identifying criminals in British India.

In 1901, Sir Henry, established the first fingerprint files in London.

Subsequently, within the next 25 years, the Henry Classification System was

adopted as the universally accepted method of personal identification by law

enforcement agencies throughout the world. It is still in use, though several

variants of the Henry Classification System exist.

In 1903, the Henry classification system was used to differentiate two

prisoners who were identical twins. The Bertillon system was not able to make

out the difference between identical twins and thus Henry classification system

was further strengthened.

Juan Vucetich also worked on a classification system based on the

findings of Sir Galton and years of experience in fingerprint forensics. His

system was published in his book, “Dactiloscopía Comparada” (Comparative

Fingerprinting) in 1904. His system, the Vucetich System, is still used in most

Spanish-speaking countries.

During the first 25 years of the 1900s, more and more agencies in the

U.S. started to send copies of their fingerprint cards to the National Bureau of

Criminal Identification. These files formed the nucleus of the FBI fingerprint

files when the Identification Division of the FBI was established in 1924. By

1946, the FBI had processed more than 100 million fingerprint cards in

40

manually maintained files. By 1971, this number had increased to 200 million

cards. The first country to adopt a national computerized form of fingerprint

imaging was Australia in 1986, which implemented fingerprint imaging

technology into its law enforcement system.

With the introduction of AFIS technology (Automated Fingerprint

Identification System), the files were split into computerized criminal files and

manually maintained civil files. Many files were found to be duplicated and the

records actually represented somewhere between 25 and 30 million criminals

and an unknown number of individuals in the civil files.

2.2.2. Fingerprint Features

While considering the various features involved with fingerprint

recognition, Galton introduced Level 2 features by defining minutiae points as

either ridge endings or ridge bifurcations on a local ridge. He also developed a

probabilistic model using minutiae points to quantify the uniqueness of

fingerprints (Galton, 1965). Although Galton discovered that sweat pores can

also be observed on the ridges, no method was proposed to utilize pores for

identification.

In 1912, Locard introduced the science of poroscopy, the comparison of

sweat pores for the purpose of personal identification (Locard, 1912). Locard

stated that like the ridge characteristics, the pores are also permanent,

immutable, and unique, and are useful for establishing the identity, especially

when a sufficient number of ridges are not available. Locard further studied the

variation of sweat pores and proposed four criteria that can be used for pore

based identification: the size of the pores, the form of the pores, the position of

the pores on the ridges and the number or frequency of the pores (Wentworth

and Wilder, 1932). It was observed that the number of pores along a centimeter

of ridge varies from 9 to 18, or 23 to 45 pores per inch and 20 to 40 pores

should be sufficient to determine the identity of a person (Ashbaugh, 1999).

41

In particular, pores provide essential information for fragmentary latent

or partial print examination since the number of minutiae points in latent

fragment prints is often too few. One such example is given in Figure 2.1a and

2.1b, where two fingerprint segments from different fingers with only one

minutia is present. The two figures show a bifurcation at the same location on

similar patterns. Normal examination would find them in agreement, but their

relative pore locations differ. Even with this small sized print, it was possible to

extract about 20 pores in these images which are sufficient to successfully

determine a match or non-match between the two prints.

Figure 2.1 : Role of Pores in Fragmentary Latent Print Examination (Source : Ashbaugh, 1999)

In 1962, Chatterjee proposed the use of ridge edges in combination with

other friction ridge formations to establish individualization, which is referred

to as “edgeoscopy”. Chatterjee discovered that some shapes on the friction

ridge edges tend to reappear frequently and classified them into eight

categories, namely, straight, convex, peak, table, pocket, concave, angle, and

others (Figure 2.2).

Figure 2.2 : Characteristic Features of Friction Ridges (Source : Ashbaugh, 1999)

42

Subsequent research established that all the edge characteristics along

friction ridges can be placed into one of these categories. It is believed that the

differences in edge shapes are caused by the effects of differential growth on

the ridge itself or a pore that is located near the edge of the friction ridge. In

theory, the density of ridge edge features can be very large, e.g., given the

average width of a ridge to be approximately 0.48 mm, a ridge 5 mm long

would contain approximately 20 edge characteristics. However, in practice, the

flexibility of the friction skin tends to mask all but the largest edge shapes.

Over the last 15 years, poroscopy and edgeoscopy have received

growing attention and have been widely studied by scientists of ridgeology, a

fundamental and essential resource for latent print examiners. It has been

claimed that shapes and relative positions of sweat pores and shapes of ridge

edges are as permanent and unique as traditional minutiae points, and, when

understood, they add considerable weight to the conclusion of identification

(Jain and Demirkus, 2007).

2.3. FINGERPRINT SENSING TECHNOLOGY

There are many different sensing methods to obtain the ridge-and-valley

pattern of finger skin or fingerprint (Xia and O’Gorman, 2003). Historically, in

law enforcement applications, fingerprints were mainly acquired offline.

Nowadays, most commercial and forensic applications accept live-scan digital

images acquired by directly sensing the finger surface with a fingerprint sensor

based on optical, solid-state, ultrasonic, and other imaging technologies.

The earliest known images of fingerprints were impressions in clay and

later in wax. Starting in the late 19th century and throughout the 20th century,

the acquisition of fingerprint images was mainly performed by using the “ink-

technique”. This kind of process is referred to as rolled offline fingerprint

sensing, which is still being used in forensic applications and background

checks of applicants for sensitive jobs.

43

Later “Live-scan” sensors with Frustrated Total Internal Reflection

(FTIR) principle were also invented. The sensors implemented earlier had the

disadvantage that they were ill-suited for wet or dry fingers and had to be

cleaned regularly to prevent grease and dirt from compromising the image

quality.

The past 15 years have envisaged tremendous development in the

fingerprint sensing technology. For example, Multispectral Fingerprint Imaging

(MSI) has been introduced by Lumidigm, Inc. (Rowe et al., 2005). Unlike

conventional optical fingerprint sensors, MSI devices scan the subsurface of

the skin by using different wavelengths of light (e.g., 470 nm (blue), 574 nm

(green), and 636 nm (red)). The fundamental idea is that different features of

skin cause different absorbing and scattering actions depending on the

wavelength of light. Fingerprint images acquired using the MSI technology

appear to be of significantly better quality compared to conventional optical

sensors for dry and wet fingers. Multispectral fingerprint images have also been

shown to be useful for spoof detection (Nixon and Rowe, 2005)

Later, in 2006, sensing technology based on multicamera system were

introduced. These were termed as “touchless imaging,” and were introduced by

TBS, Inc. (Parziale and Diaz-Santana, 2006). As suggested by the name,

touchless imaging avoids direct contact between the sensor and the skin and,

thus, consistently preserves the fingerprint “ground truth” without introducing

skin deformation during image acquisition. A touchless fingerprint sensing

device is also available from Mitsubishi (http://global.mitsubishielectric.com,

2006).

One of the most essential characteristics of a digital fingerprint image is

its resolution, which indicates the number of dots or pixels per inch (ppi).

Generally, 250 to 300 ppi is the minimum resolution that allows the feature

extraction algorithms to locate minutiae in a fingerprint image. FBI-compliant

sensors must satisfy the 500 ppi resolution requirement. However, in order to

44

capture pores in a fingerprint image, a significantly higher resolution (≥ 1,000

ppi) of image is needed.

Although it is not yet practical to design solid-state sensors with such a

high resolution due to the cost factor, optical sensors with a resolution of 1,000

ppi are available commercially. More excitingly, optical sensors with

resolutions of 4,000-7,000 ppi have also been developed, which not only allow

capturing Level 3 features for identification, but also pore activities (opening

and closing) for spoof detection.

Recent years have seen a new high resolution fingerprint device called

P3400. This is a small and cost effective fingerprint reader introduced by

Zvetco Inc. (http://www.zvetcobiometrics.com). This device can produce 500

dpi images and is constructed of high-quality aluminum. It is equipped with a

6-foot USB cable and is compatible with most biometric security access

software packages.

The compact Guardian (www.crossmatch.com/l-scan-guardian.php)

includes features such as patented auto capture capability and Perfect Roll

technology, making it ideal for foolproof fingerprint acquisition in high-volume

processing environments, such as visa issuance and border control. The device

can collect quality fingerprints at high resolution (500 dpi) in few seconds and

meets international standards that comply with government requirements in

many countries.

The ten-print MFS-500 live scanner is high resolution (500 and

1000dpi) devices designed and built for optical perfection. It can read plane

static fingerprints and can also be applied in 3- D print-pressed rolling scan. A

sensor provides very clear image and prevents dry fingerprint problems. A

variety of fingerprint identification software is added into the sensor to identify

the fingerprint. It is claimed by the manufacturers that the fingerprint captured

with this scanner can have the highest identification and matching ratio

45

(http://www.mantratec.com/ biometrics / SLAP-4-4-2- fingerprint - scanner -

forensic-biometrics-MFS-500.html).

Futronic FS80 USB2.0 Fingerprint Scanner (www.bayometric.com)

uses an advanced CMOS sensor technology and precise optical system to

deliver high quality fingerprint image. The finger is illuminated by 4 infra-red

LED's during scanning and the light intensity is automatically adjusted

according to scanning fingerprint's characteristics (wet, dry, blurred, etc.) to

optimize the quality of the captured fingerprint image. It captures an un-

distorted raw fingerprint image of 500dpi resolution into PC in 100msec. The

scanner can reject fake fingers made from silicone rubber and play-doh. It

supports fingerprint recognition, verification, authentication, and matching

applications.

The type of scanner used depends upon the application and environment

where it is to be implemented. In general, it is preferred to have scanners that

are products certified for compliance with the FBI's integrated automated

fingerprint identification system image quality specifications. These

specifications include details regarding fingerprint image resolution, size

(area), number of pixels, geometric accuracy, gray-level quantization and gray

range, spatial frequency response and signal-to-noise (SNR) ratio. The scanners

certified by the FBI as tested and in compliance with the FBI's Next Generation

Identification (NGI) initiatives and Integrated Automated Fingerprint

Identification System (IAFIS) Image Quality Specifications (IQS) as of 2012 is

presented in https://www.fbibiospecs.org/IAFIS/Default.aspx.

2.3.1. Fingerprint Acquisition Methods

This section presents the various acquisition methods used to obtain

fingerprints of an individual.

46

• Optical

Optical fingerprint imaging involves capturing a digital image of the

print using visible light. This type of sensor is, in essence, a specialized digital

camera. The top layer of the sensor, where the finger is placed, is known as the

touch surface. Beneath this layer is a light-emitting phosphor layer, which

illuminates the surface of the finger. The light reflected from the finger passes

through the phosphor layer to an array of solid state pixels (a charge-coupled

device), which captures a visual image of the fingerprint. A scratched or dirty

touch surface can cause a bad image of the fingerprint. A disadvantage of this

type of sensor is the fact that the imaging capabilities are affected by the

quality of skin on the finger. For instance, a dirty or marked finger is difficult

to image properly. Also, it is possible for an individual to erode the outer layer

of skin on the fingertips to the point where the fingerprint is no longer visible.

It can also be easily fooled by an image of a fingerprint if not coupled with a

“live finger” detector. However, unlike capacitive sensors, this sensor

technology is not susceptible to electrostatic discharge damage (Raul, 2007).

• Ultrasonic

Ultrasonic sensors make use of the principles of medical

ultrasonography in order to create visual images of the fingerprint. Unlike

optical imaging, ultrasonic sensors use very high frequency sound waves to

penetrate the epidermal layer of skin. The sound waves are generated using

piezoelectric transducers and reflected energy is also measured using

piezoelectric materials. Since the dermal skin layer exhibits the same

characteristic pattern of the fingerprint, the reflected wave measurements can

be used to form an image of the fingerprint. This eliminates the need for clean,

undamaged epidermal skin and a clean sensing surface (Majid and Saeed,

2005).

47

• Capacitance

Capacitance sensors utilize the principles associated with capacitance to

form fingerprint images. In this method of imaging, the sensor array pixels

each act as one plate of a parallel-plate capacitor, the dermal layer (which is

electrically conductive) acts as the other plate, and the non-conductive

epidermal layer acts as a dielectric (Setlak, 2005).

• Passive capacitance

A passive capacitance sensor uses the principle outlined above to form

an image of the fingerprint patterns on the dermal layer of skin. Each sensor

pixel is used to measure the capacitance at that point of the array. The

capacitance varies between the ridges and valleys of the fingerprint due to the

fact that the volume between the dermal layer and sensing element in valleys

contains an air gap. The dielectric constant of the epidermis and the area of the

sensing element are known values. The measured capacitance values are then

used to distinguish between fingerprint ridges and valleys (Setlak, 2005).

• Active capacitance

Active capacitance sensors use a charging cycle to apply a voltage to the

skin before measurement takes place. The application of voltage charges the

effective capacitor. The electric field between the finger and sensor follows the

pattern of the ridges in the dermal skin layer. On the discharge cycle, the

voltage across the dermal layer and sensing element is compared against a

reference voltage in order to calculate the capacitance. The distance values are

then calculated mathematically, and used to form an image of the fingerprint.

Active capacitance sensors measure the ridge patterns of the dermal layer like

the ultrasonic method. Again, this eliminates the need for clean, undamaged

epidermal skin and a clean sensing surface (Setlak, 2005). A comparison of

optical and non-optical live scan fingerprint scanners is given in Table 2.1.

48

TABLE 2.1

COMPARISON OF OPTICAL AND NON-OPTICAL SENSORS

Optical Non-optical

Measuring

Method

Light Pressure, heat, capacitance,

ultrasonic wave

Strength Highly-stable performance

physical / electrical durability

high-quality image

Low cost with mass

production, compact size

integrated with low-power

application

Weakness Relatively high cost

limit to size-reduction

relatively easy to fool with a

finger trace or fake finger

Physical/electrical weakness

performance sensitive to the

outer environment

(temperature, dryness of a

finger)

Application Entrance time and attendance

control, banking service, PC

security

PC security, e-commerce,

authentication, mobile devices

and smart cards

2.4. FINGERPRINT ENHANCEMENT TECHNIQUES

Fingerprint enhancement is a preprocessing technique to make the

image clearer than the original image, for further operations. In other words,

enhancement techniques improve the quality of the fingerprint images. The

quality is determined by many factors, which sometime may be difficult to

control, therefore a fingerprint system must be able to handle also the images of

medium and low quality (recoverable). In some cases, it is possible to improve

significantly the image quality by applying some image enhancement

technique. The main purpose of such procedure is to enhance the image by

improving the clarity of ridge structure or increasing the consistence of the

ridge orientation. In noisy regions, it is difficult to define a common orientation

of the ridges. The process of enhancing the image before the feature extraction

is also called pre-processing. The reason for degradation lies in the fact that

images acquired from sensors or other media are not always assured of perfect

quality. Several types of filters in both spatial and frequency domains have

been proposed in the literature. This section reviews some of these works.

49

The basic fingerprint image preprocessing operations are:

(i) Noise Reduction and

(ii) contrast enhancement which enhances the contrast between

darker and brighter curves in a fingerprint image.

There appear many algorithms and techniques proposed and applied to

fingerprint image enhancement. The main algorithms use Fourier transform

(Sherlock et al., 1994; Chikkerur and Govindaraju, 2005), Gabor filters,

Wavelet transform (Hsieh et al., 2003; Paul and Lourde, 2006; Ye et al., 2007;)

and minutiae filtering, applied to binary (Farina et al., 1999) or gray-scale

images (Maio and Maltoni, 1997).

One of the most widely cited fingerprint enhancement techniques is the

method employed by Hong et al. (1998), which is based on the convolution of

the image with Gabor filters tuned to the local ridge orientation and ridge

frequency. Gabor filters (Gabor, 1946) have both frequency-selective and

orientation-selective properties and have optimal joint resolution in both spatial

and frequency domains. Therefore, it is beneficial to use Gabor filters as

bandpass filters to remove the noise and preserve true ridge/valley structure.

The stages of this algorithm include normalisation, ridge orientation estimation,

ridge frequency estimation and filtering. This method was a follow up of their

previous work (Hong et al., 1996). In this work, the authors introduced a new

fingerprint enhancement algorithm that decomposes the input fingerprint image

into a set of filtered images. A set of band pass filters can efficiently remove

the undesired noise and preserve the true ridge/valley structure.

Yang et al. (2003) modified the method proposed by Hong et al. (1998)

by discarding the inaccurate prior assumption of sinusoidal plane wave, and

making the parameter selection process independent of fingerprint image.

Greenberg et al. (2000) proposed the use of an anisotropic filter that

adapts its parameters to the structure of the underlying sub-region. Wu et al.

50

(2004) proposed to convolve a fingerprint image with an anisotropic filter to

remove Gaussian noise and then apply Directional Median Filter (DMF) to

remove impulse noise. On visual inspection, the enhancement results of Wu et

al. (2004) appear to be superior to those obtained by Greenberg et al. (2000).

Sherlock et al. (1994) proposed a fingerprint enhancement method in the

Fourier domain. In this approach, a fingerprint image is convolved with pre-

computed filters, which result in a set of filtered images. The enhanced

fingerprint image is constructed by selecting each pixel from the filtered image

whose orientation is the closest to that of the original pixel.

Teddy and Martin (2002) described the latent fingerprint image

enhancement using spectral analysis technique. The latent fingerprints are often

blurred, incomplete, degraded and their spatial definition is not clear. This

paper has presented techniques from frequency (spectral) analysis that can be

used for the enhancement and restoration of degraded, noisy and sometimes

incomplete fingerprint by using high-pass Butterworth filter and/or band-pass

Butterworth filter. Rolled or flat fingerprint captured using ink or live scan

usually need only the spatial filtering techniques, such as brightness, contrast,

gamma and/or color map adjustment to examine the minutiae information.

However for latent fingerprint, besides the spatial image enhancement filtering,

one needs to use frequency (spectral) analysis techniques or a combination of

both spatial and frequency enhancement techniques to isolate and enhance the

degraded and often very weak, fingerprint information from a variety of

background patterns.

Chikkerur (2005) proposed an algorithm based on Short Time Fourier

Transformation (STFT), and a probabilistic approximation of dominant ridge

orientation and frequency was used instead of the maximum response of the

Fourier spectrum. The ridge orientation image, ridge frequency image and

foreground region image are generated simultaneously while performing the

STFT analysis.

51

E-Kyung and Bae (2006) proposed an adaptive filter according to

oily/dry/neutral images, instead of uniform filtering. To identify

oily/dry/neutral, five features such as Mean, Variance, Block directional

difference, Ridgevalley thickness ratio and Orientation change are used for

clustering by Ward’s clustering algorithm. After clustering, if the image is dry,

then the ridges are enhanced by extracting their centerlines and removing white

pixels, that is, ridge enhancement. For oily images, valleys are enhanced by

dilating thin and disconnected ones, that is, 0valley enhancement. For neutral

image, there is no need for filtration.

Recently, Chengpu et al. (2008) proposed the enhancement technique by

using the combination of Gabor filter and Diffusion filter methods. The authors

have combined the advantages of Gabor filtering and Diffusion filtering

methods and proposed an enhancement method using the two filters: the low-

pass filter (1D Gaussian filter) and the band-pass filter (1D Gabor filter).

2.5. SEGMENTATION TECHNIQUES

Before extracting the feature of a fingerprint, it is important to separate

the fingerprint regions (presence of ridges) from the background. This limits

the region to be processed, and therefore, reduces the processing time and false

feature extraction. A correct segmentation may be, in some cases, very

difficult, especially in poor quality fingerprint image or noisy images, such as

presence of latents. The same information used for quality extraction, such as

the contrast, ridge orientation and ridge frequency can be used for the

segmentation or inclusive of the quantified region quality may be used directly

by considering as background, the regions with quality below some threshold.

Normally, the segmentation is also computed by block in the same way as the

quality extraction.

Segmentation is considered as a critical step in automatic fingerprint

recognition. There is a large literature on segmentation dating back over 30

years (Jain and Dubes, 1988). Some previous approaches to image

52

segmentation, which provide the basis for more recent methods, include

boundary-based segmentation such as Canny edge detection (Canny, 1986),

region-based segmentation such as region growing (Adams and Bishof, 1994;

Chakraborty et al., 1996) and global optimization approaches such as those

based on Mumford-Shah functional (Klingler et al., 1988; Chakraborty et al.,

1996). The discussion of this section is divided into two parts. The first

presents some techniques that are general in nature and can be applied to any

type of images, while the second section presents works focused on segmenting

fingerprint from background.

Several approaches to fingerprint image segmentation are known from

literature. In early days, Mehtre et al. (1987) implemented a segmentation

algorithm that first partitioned a fingerprint image into blocks of 16 16

pixels. Then, each block was classified according to the distribution of the

gradients in that block. Later Mehtre and Chatterjee (1989) extended this

method by excluding blocks with a gray-scale variance that is lower than some

threshold.

The gray-scale variance in the direction orthogonal to the orientation of

the ridges was used to classify each 16 16 block by Ratha et al. (1995). Jain

and Ratha (1997) used the output of a set of Gabor filters as input to a

clustering algorithm that constructs spatially compact clusters. Sun and Ai

(1996) used dynamic threshold value (T) to binarize the input fingerprint

images, where the dynamic threshold was suggested by Moayer and Fu (1975).

The method used 5x5 pixels neighborhood to determine the local threshold

value (T), which is equal to the mean value of the 5x5 pixels.

The segmentation algorithm proposed by Bazen and Gerez (2000) was

based on the coherence and morphology to obtain smooth regions. The same

authors (Bazen and Gerez, 2001) enhanced their work to two more features

namely mean and variance along with coherence. An optimal linear classifier is

trained for the classification per pixel, while morphology is applied as post

53

processing to obtain compact clusters and to reduce the number of

classification errors.

Naji et al. (2002) proposed a segmentation algorithm based on

histogram equalizer and automated the method of choosing the threshold value

during segmentation.

Segmentation algorithms can generally be unsupervised, where a

threshold is set on detected features to segment the image, or supervised where

a simple linear classifier is used to classify features as part of region of interest

or the background. Examples of supervised methods include the work of

Alonso-Fernandez et al. (2005), where Gabor filters were used for

segmentation. Apart from supervised and unsupervised methods, neural

networks are also used. Barreto et al. (2005) train a neural network on the

Fourier spectrum of fingerprint images to segment the image. Similarly, Zhu

et al. (2006) utilized a neural network trained on the gradient of the fingerprint

orientation to segment the images. Wu et al. (2007) proposed an approach

based on strength of Harris corners in the image, which is used to extract the

region of interest. This approach used the corner strength measure to

distinguish between the region of interest and the background.

Segmentation process can become very complex and intricate because

the boundary between the region of interest and the background blurs due to

the presence of noise. Various segmentation methods are developed. But these

methods are not absolutely satisfying. For example, if there is a highly noisy

background region surrounding the poor contrast foreground of the fingerprint

image, these methods will fail to separate background with foreground. A

robust segmentation method is required to deal with low quality images and to

be insensitive to the contrast of the original images.

54

2.6. MATCHING ALGORITHMS

Fingerprint matching refers to finding the similarity between two given

fingerprint images. The choice of the matching algorithm depends on which

fingerprint representation is being used. Typically, a matching algorithm first

attempts to recover the translation, rotation and deformation parameters

between the given image pair and then determines the similarity between the

two images.

Fingerprint matching is considered a challenging problem due to the

noise in the fingerprint images, large intra-class variation and small interclass

variations between different impressions of the same finger. As each

authentication application has different performance requirement, there is a

need to continually improve the matching performance of the current systems.

This section reviews some of the reported matching algorithms.

The available matching algorithms can be broadly classified into five

categories depending on the type of features used. They are

• Correlation based matching

• Minutia based matching

• Ridge feature based matching

• Non-minutia based matching

• Hybrid Methods

2.6.1. Correlation Based Matching

In order to match two fingerprints using the correlation based technique,

the fingerprints are aligned and the correlation is computed for each

corresponding pixels, however, as the displacement and rotation are unknown it

is necessary to apply the correlation for all possible alignments. The singularity

information may be useful in order to find an approximated alignment. The

main drawback of this method is its computational complexity and less

tolerance to non-linear distortion and contrast variation.

55

There have been some alternative proposals that compute the correlation

locally instead of globally, in which only interesting regions (e.g., minutia and

singularity regions) are selected and matched. These algorithms use simple

techniques to align two fingerprint images and subtract the input image from

the template image to see if the ridges correspond. This method has several

disadvantages.

a. It fails if the images are highly distorted. The distortion is more

pronounced in global fingerprint patterns; thus considering the local

regions can minimize distortion to some extent. Bazen et al. (2000) and

Nandakumar and Jain (2004) present some approaches to localized

correlation-based matching.

b. Another factor that affects the performance of correlation-based

matching algorithms is the variations introduced through finger

pressure, skin condition, image brightness, contrast and ridge thickness

of the same finger. When these situations arise, a more sophisticated

correlation measures such as normalized cross-correlation or zero-mean

normalized cross-correlation are needed. These techniques can be used

to compensate contrast and brightness variations and the application of

enhancement, binarization and thinning steps may limit the ridge

thickness problem (David et al., 2005).

c. Increased complexity : The computational complexity of this method is

high. This problem can be solved by using Fourier domain method

(Coetzee and Botha, 1993) and Fourier-Mellin transformation (Sujan

and Mulqueen, 2002).

2.6.2. Minutia Based Methods

This is the most popular and is widely used in commercial applications,

because of its good performance and low computation time, especially for good

quality images. This method tries to align the minutiae of the input image

56

(query template) and stored templates (reference template) and find the number

of matched minutiae. After alignment, two minutiae are considered in matching

if the spatial distance and direction difference between them are smaller than a

given tolerance. A correct aligning of fingerprint is very important in order to

maximize the number of matched minutiae; this requires the computing of the

translation and rotation information, as well as other geometrical

transformations such as scale and distortion. In order to compute efficiently

aligning information, there have been many approaches proposed.

Fingerprint alignment or registration is a crucial step in fingerprint

recognition. Its goal is to retrieve the transformation parameters between

fingerprint images and then align them for matching. Some non-rigid

deformation or distortion could occur in fingerprint image acquisition. It is very

costly to model and remedy such distortions in fingerprint registration, and they

can be compensated to some extent in subsequent fingerprint matching. Thus, a

majority of existing fingerprint alignment methods consider only translation

and rotation, although some deformable models (Cappelli et al., 2001; Ross et

al., 2005) have been proposed.

According to the features used, the existing fingerprint alignment

methods can be divided into two categories, minutia based and non-minutia

feature based methods. Minutia based methods are now the most widely used

ones (Nanni and Lumini, 2008). Non-minutia feature based methods (Jain et

al., 2000; Zhang and Wang, 2002), include those using image intensity values,

orientation fields, cores, etc. One problem in applying these methods to partial

fingerprints is that the features required by them could be very few on the

fragments. Consequently, they will either lead to incorrect results or be not

applicable.

There are roughly two kinds of methods for estimating alignment

transformations. The first kind of methods quantizes the transformation

parameters into finite sets of discrete values and searches for the best solution

57

in the quantized parameter space (Ratha et al., 1996; Chang et al., 1997; Bazen

et al., 2000; Ross et al., 2002; Yager and Amin, 2005, 2006; Kryszczuk et al.,

2004a, 2004b; Liu et al., 2006). The alignment accuracy of these methods is

thus limited due to the quantization.

The second kind of methods first detects corresponding feature points

(or reference points) on fingerprints and then estimates the alignment

transformation based on the detected corresponding points (Jain et al., 1997;

Huvanandana et al., 2000; ; Jiang and Yau, 2000; Kovacs-Vajna, 2000; Jain et

al., 2000; Zhang and Wang, 2002; Tico and Kuosmanen, 2003; Chen et al.,

2006a, 2006b; Feng, 2008). Most of such methods make use of minutiae as the

feature points. As discussed before, however, it is problematic to align partial

fingerprints based on minutiae because of the lack of such features on the

fingerprint fragments.

The minutia-based methods use minutiae points as features and

represent different attributes such as location, orientation, type, quality of the

neighbourhood region, etc. Minutia-based techniques first find minutiae points

and then map their relative placement on the finger. However, there are some

difficulties when using this approach. It is difficult to extract the minutiae

points accurately when the fingerprint is of low quality. Also, this method does

not take into account the global pattern of ridges and furrows. Further,

fingerprint matching based on minutiae has problems in matching different

sized (unregistered) minutiae patterns. Local ridge structures cannot be

completely characterized by minutiae.

The minutia based method can further be classified as Global matching

(Ranade and Rosenfeld, 1980) and Local matching methods. The global

matching approaches try to simultaneously align all the minutiae points. The

alignment methods can be either implicit or explicit. The implicit method tries

to find the point correspondences, and, in the process, an optimal alignment is

obtained. The explicit algorithm aligns the minutiae sets first and then finds the

58

point of correspondences. Local matching approaches try to match local

minutiae structures, which are characterized by attributes that are invariant with

respect to global transformation.

The local versus global matching is a trade-off among simplicity, low-

computational complexity, high distortion tolerance (local matching) and high

distinctiveness (global matching). Local matching methods are more robust to

non-linear distortion and partial overlaps when compared to global approaches.

Matching local minutiae structures relax global spatial relationships, which are

considered to be highly distinctive, and therefore, reduce the amount of

information available for discriminating fingerprints.

Examples of global matching approaches include the use of Hough

transform (Stockman et al., 2000), hierarchical transform (Karu and Jain, 1996)

and adaptive elastic string matching algorithm (Cormen et al., 2001). Similarly,

examples of local matching approaches include Jiang and Yau (2000), Ratha et

al. (2000), Cormen et al. (2001) and Chikkerur et al. (2006).

2.6.3. Ridge Feature Based Matching

A matching using the ridge feature in form of finger code consists of

computing the difference of two finger code vectors (query and reference).

However, before applying the finger code, it is important to align the

fingerprint images, which is really a big problem, as in the case of other

methods. In some cases the singularity may be used for that purpose. A finger

code also may be used as a complementary to minutia based method in order to

improve the overall matching accuracy. The original approach of this method

used circular finger codes, considering as center the core point. The final result

of the finger code difference is normalized and averaged using the eight

directions and obtained a value that varies from 0 to 1. The lower the score, the

more similar are the fingerprints. Some threshold values are used to decide

whether there is matching or not.

59

2.6.4. Non-Minutia Based features

As mentioned previously, minutiae points have several disadvantages

and therefore alternative features are being examined. The popular alternative

features are

(i) Global and local texture information and

(ii) Level 3 features.

Global and local texture information are important alternatives to

minutiae. Textures are defined by spatial repetition of the basic elements and

are characterized by properties such as scale, orientation, frequency, symmetry,

and isotropy. Fingerprint ridge lines are mainly described but smooth ridge

orientation and frequency, except at singular regions. These singular regions

are discontinuities in a basically regular pattern and include the loops and

deltas at a coarse resolution and the minutiae points at a higher resolution.

Global texture analysis fuses contributions from different characteristic regions

into a global measurement and as a result, most of the available spatial

information is lost. Local texture analysis has proved to be more effective than

global feature analysis (Jain et al., 1999).

Hybrid methods that combine texture and minutiae features have also

been proposed (Nanni and Lumini, 2008). This method used Local Binary

Patterns (LBP)as fingerprint descriptors. In this system, two fingerprints to be

matched are first aligned using their minutiae, then the images are decomposed

in several overlapping sub-windows, each sub-window is convolved with a

bank of Gabor filters and, finally, the invariant local binary patterns histograms

are extracted from the convolved images.

Nikam and Agarwal (2008) also used LBP features along with wavelets

for fingerprint detection. Local binary pattern (LBP) histograms are used to

capture these textural details. Wavelet energy features, characterizing ridge

frequency and orientation information, are also used for improving the

60

efficiency of the proposed method. Dimensionalities of the feature sets are

reduced by running sequential forward floating selection (SFFS).

Existing matching algorithms can be classified into two categories,

global and local feature-based algorithms. Global feature-based algorithms aim

at recognizing an object as a whole. This class of algorithm is suitable for

recognition of homogeneous (textureless) objects, which can be easily

segmented from the image background. Examples include Hu moments

(Vuppala et al., 2007) and the eigenvectors of the covariance matrix of the

segmented object (Lee et al., 2005). Global features based Recognizers are

simple and fast but there are limitations in the reliability of object recognition

under variations. In contrast to this, local features based algorithms are more

suitable for textured objects and are more robust with respect to variations. The

advantages of local over global features are demonstrated by Ke et al. (2004).

Local features based algorithms focus mainly on the so-called keypoints.

In this context, the general scheme for object recognition usually involves three

important stages: The first one is the extraction of salient feature points (for

example corners) from both the test and model object mechanisms that aim to

keep the region characteristics insensitive to viewpoint and illumination

changes. The final stage is the matching between test and model images based

on the extracted features.

The development of image matching by using a set of local keypoints

can be traced back to the work of Moravec (1977). He defined the concept of

“points of interest” as being distinct regions in images that can be used to find

matching regions in consecutive image frames. The Moravec operator was

further developed by Harris and Stephens (1998) who made it more repeatable

under small image variations and near edges. Schmid and Mohr (1997) used

Harris corners to show that invariant local features matching could be extended

to the general image recognition problem. They used a rotationally invariant

descriptor for the local image regions in order to allow feature matching under

61

arbitrary orientation variations. Although it is rotational invariant, the Harris

corner detector is however very sensitive to changes in image scale, and

therefore, does not provide a good basis for matching images of different sizes.

Lowe (1999; 2001; 2004) overcame such problems by detecting the

points of interest over the image and its scales through the location of the local

extrema in a pyramidal Difference of Gaussians (DOG). The Lowe’s

descriptor, which is based on selecting stable features in the scale space, is

named the Scale Invariant Feature Transform (SIFT). Mikolajczyk and Schmid

(2005) experimentally compared the performances of several currently used

local descriptors and they found the SIFT descriptors to be the most effective,

as they yielded the best matching results.

It is only in the past few decades that SIFT is being used in fingerprint

recognition and verification. Yu and Huang (2009) proposed a SIFT-based

image fingerprinting approach that is robust to geometric transformations. The

algorithm using the SIFT features utilized a method based on area ratio

invariance of affine transformation to verify valid matched key point pairs

between the queried image and the pre-registered image. By counting the valid

matched pairs, they estimated whether the two images are homologous or not.

Park (2008) demonstrated the utility of SIFT representation for

fingerprint-based identification. In this, a map of gradient orientations was

generated around each local extremum and then the descriptor orientation

invariant was made. All gradient orientations are rotated with respect to the

major orientation (keypoints orientation) of the local extremum.

Iannizzotto and Rosa (2008) proposed another method that uses SIFT

features for fingerprint verification system. The aim was to re-formulate an

algorithm for fingerprint verification using SIFT to exploit the high degree of

parallelism inherent in a single-layer Cellular Neural Network.

62

2.6.5. Hybrid Methods

Recently, researchers have come up with hybrid fingerprint matchers by

making use of more than one basic approach to matching. For example, Ross

et al. (2003) have suggested the use of both minutiae and ridge flow

information to represent and match fingerprints. They have shown that the

performance of the minutia-based matcher presented by Jain et al. (1997) can

be significantly improved by using additional information provided by the

Finger Code method (Jain et al., 2000).

The local correlation-based fingerprint matching algorithm presented in

this paper is a similar attempt to improve the performance of a minutia-based

matcher by introducing a correlation step to ascertain the quality of each

minutia match. The gray-level information of the pixels around the minutiae

points contains richer information about the local region than the attributes of

the minutiae points. Hence, the spatial correlation of regions around

corresponding minutiae points is a good measure of the degree of similarity

between them.

The correlation-based fingerprint matcher proposed by Bazen et al.

(2000) selects certain distinctive regions in the template fingerprint image and

searches for those regions in the query image. However, their method is not

very robust to rotation. Beleznai et al. (2001), in order to improve the

recognition performance of a minutia-based matching system, exploit the

structural information around minutiae.

2.7. LEVEL 3 FEATURES BASED RECOGNITION

The use of Level 3 features in an automated fingerprint identification

system has been studied by only a few researchers. There is a growing interest

in using Level 3 features, such as pores (Stosz and Alyea, 1994; International

Biometric Group, 2008), ridge contours (Jain et al., 2007) and, dots and

incipient ridges (Chen and Jain, 2007), for fingerprint matching. It is claimed

63

that Level 3 features contain discriminating information and can improve the

performance of matching rolled/plain to rolled/plain fingerprints. This section

reviews works that focus on the extraction of pores and matching performance

from high resolution images.

A critical step in the pore based AFRS is the extraction of pores from

fingerprint images. The existing methods extract pores by using skeleton

tracking-based or filtering-based approaches. The skeleton-tracking-based

approaches are quite time- consuming and work well only with very high

quality fingerprint images. The filtering-based approaches are more efficient

and more robust. They use static isotropic pore models to detect pores.

2.7.1. Skeleton Tracking-based Approaches

Stosz and Alyea (1994) proposed a skeletonization-based pore

extraction and matching algorithm. Specifically, the locations of all end points

(with at most one neighbor) and branch points (with exactly three neighbors) in

the skeleton image are extracted and each end point is used as a starting

location for tracking the skeleton. The tracking algorithm advances one

element at a time until one of the following stopping criteria is encountered:

1) another end point is detected

2) a branch point is detected

3) the path length exceeds a maximum allowed value.

Condition 1 implies that the tracked segment is a closed pore, while Condition

2 implies an open pore. Finally, skeleton artifacts resulting from scars and

wrinkles are corrected and pores from reconnected skeletons are removed. The

result of pore extraction is shown in Figure 2.3.

64

Figure 2.3 : Detected Pore Locations denoted by Square Box

During matching, a fingerprint image is first segmented into small

regions and those that contain characteristic features, such as core and delta

points, are selected. The match score between a given image pair is then

defined as a ratio of the number of matched pores to the total number of pores

extracted from template regions, and is given by,

��

�

�

��

�

��

−

=

��

�

�

��

�

��

−

=

=

1sN

0ii,PN

1sN

0ii,MPN

pS (2.1)

where Ns is the total number of regions in the template, NP,i is the number of

pores detected in template region i and NMP;i is the number of matching pores

in region i. Here, alignment is first established based on maximum intensity

correlation and two pores are considered matched if they lie within a certain

bounding box. Experiment results published were obtained by using a database

of 258 fingerprints taken from 137 individuals by combining minutia and pore

information. A lower FRR of 6.96 percent (compared to 31 percent for

minutiae alone) at a FAR of 0.04 percent was obtained.

Later, Roddy and Stosz (1997) conducted a statistical analysis of pores

and presented a model to predict the performance of a pore-based automated

fingerprint system. In general, this study provided statistics about pores and

65

demonstrates the efficacy of using pores, in addition to minutiae, for improving

the fingerprint recognition performance.

Kryszczuk et al. (2004a, 2004b) conducted research to find the

advantage of level 3 features over level 2 features in fingerprint fragment

matching when supplied with high-resolution images. In order to extract pores,

first the image is binarised by applying a threshold method. In the binary

image, closed pores look like a ‘hole’ in a ridge. They focus on areas of white

pixels entirely surrounded by black pixels. Then the upper and lower thresholds

tmin and tmax were defined. Each suspected pore whose surface is below tmin and

above tmax is discarded. The object of a surface smaller than tmin and larger than

tmax are removed, because they are noise artifacts or not likely to be pore

feature. The next step is to extract open pore look like ‘hook’ and skeletalise

the valleys of binarised image. Having done so, the spurs appear at pore

location. The distance between the end of each spur and skeleton of valleys are

computed. If this distance is sufficiently large and amount of white pixels

around the spur location is big enough, then the end of the spur was consider as

pore location. They used 2000 dpi images using a custom built scanner and the

database had 12 genuine and 6 impostor images. Comparison was performed

by extracting pores and ridges in conjunction with minutiae points, using

geometric distance criterion. According to their findings, the advantage of level

3 features can be envisaged when the size of the partial fingerprint decreases

and the number of level 3 features extracted from a high quality partial image is

the same as the number of level 2 features obtained from a full image.

A hierarchical matching system that utilized level 1, 2 and 3 features

from 1000 dpi scanned fingerprint images was proposed by Jain et al. (2006;

2007). The level 3 features extracted were pores and ridge contours that were

locally matching in windows associated with matched minutiae points using an

algorithm called Iterative Closest point (ICP) (Besl and McKay, 1992). The

experiments showed that a relative reduction of 20% occurred in EER, while

combining level 3 features with level 1 and 2 features.

66

Vatsa et al. (2009) used a score-level fusion technique and combined

level 2 and level 3 match scores to improve fingerprint recognition accuracy.

The match scores were obtained using classifiers which were first augmented

with a quality score that was determined using redundant discrete wavelet

transform of an image. These match scores were then fused using Dezert-

Smarandache theory. Experimental results showed that the performance of the

fusion method was high when compared with the existing fusion techniques

and the system worked well in presence of imprecise, inconsistent and

incomplete fingerprint information.

2.7.2 Filtering-Based Methods

Skeleton-tracking-based methods first binarize and skeletonize the

fingerprint image and then track the fingerprint skeletons. A pore is detected

when certain criteria are met during the tracking. However, skeletonization is

computationally expensive and very sensitive to noise and it works well only

on very high resolution fingerprint images of high quality. For example, the

fingerprint images used by Stosz and Alyea (1994), Roddy and Stosz (1997),

Kryszczuk et al. (2004a and 2004b) are all at least 2000 dpi.

Recently proposed approaches are filtering-based methods that detect

pores by using pore models to filter fingerprint images. Figure 2.4 shows three

typical isotropic pore models: Ray’s model (Ray et al., 2005), Jain’s model

(Jain, 2006, 2007), and the DoG (Difference of Gaussian) model (Parsons et

al., 2008).

Figure 2.4 : Three Typical Pore Isotropic Models (a) Ray’s model (Ray et al., 2005) (b) Jain’s model (Jain, 2006; 2007) (c) DoG model (Parsons et al., 2008)

67

Ray et al. (2005) proposed an approach to extract pores from fingerprint

images based on the pore model in Figure 2.5(a), which is a modified 2-

dimensional Gaussian function. They first calculated an error map for the

fingerprint image, with each entry in this map being the sum of the squared

errors between the pore model and the local area surrounding the pixel. The

error map is then binarized such that only the areas of high pore probability

(i.e. low error) are retained. In these areas, the pores are detected as the local

minima in a (2rm) x (2rm) neighborhood. The system used unitary parameters r

(the variance of the Gaussian) and rm to detect pores. However, the pore scales

and ridge/valley widths could vary greatly from one fingerprint to another

fingerprint or from one region to another region in the same fingerprint (Figure

2.5). In the figure, the left image has a closed pore while the right image has

two open pores.

Figure 2.5 : Fingerprint Images with Different Ridge and Valley Widths

Moreover, Ray’s pore model is isotropic, yet from the figure, it can be

seen that the appearance of open pores on real fingerprint images is not

isotropic. Jain et al. (2006; 2007) proposed Mexican hat wavelet transform to

extract pores based on the observation that pore regions typically have a high

negative frequency response, as intensity values change abruptly from bright to

dark at the pores. The Mexican hat wavelet actually serves as the pore model,

and its scale parameter is experimentally set for specific datasets. Figure 2.4(b)

shows the Mexican hat wavelet. Obviously, it is also isotropic. This pore model

68

is also limited in that it cannot adapt itself to different fingerprints or different

regions on a fingerprint.

Another pore extraction method was proposed by Parsons et al. (2008).

Its basic idea is to use a band-pass filter to detect circle-like features. In other

words, the method assumes that pores appear as circular objects on fingerprint

images, and the pore is thus modeled by the DoG filter. Figure 2.4(c) shows

this pore model. This method did not consider the variation of pore scales in

fingerprint images but simply used a unitary scale in their model.

To deal with the limitations caused by the unitary scale, an adaptive

DoG-based pore extraction method was proposed by Zhao et al. (2008). It

divides a fingerprint image into blocks and defines for each block, a DoG filter

according to the local ridge period on the block. One limitation of the DoG-

based methods is that the pore models are isotropic. The underlying assumption

that pores are circular features does not hold well on real fingerprint images.

To solve the above issues, Zhao et al. (2010a; 2010b) showed that

instead of using the computationally expensive skeletonization method or scale

isotropic method for pore extraction, a simple adaptive anisotropic pore model

will prove to be efficient. In this method, the fingerprint image was partitioned

into blocks and a local pore model was determined for each block. With this

model, a matched filter was used to extract the pores within each block.

Experimental results using high resolution (1200 dpi) fingerprint images

demonstrated that the proposed pore model and the pore extraction method

could locate pores more accurately and robustly in comparison with other state-

of-the-art pore extractors.

Although some methods have been proposed to construct full fingerprint

templates from a number of partial fingerprint images (Choi et al., 2007), it is

expensive or even impossible to collect sufficient fingerprint fragments to

construct a reliable full fingerprint template. Moreover, some errors (e.g.,

spurious features) could be introduced in the construction process. Thus, it is

69

meaningful and very useful if algorithms can be developed for aligning and

matching partial fingerprints to partial fingerprints.

Some researchers have studied the problem of matching a partial

fingerprint to full template fingerprints. Jea and Govindaraju (2005) proposed a

minutia-based approach to matching incomplete or partial fingerprints with full

fingerprint templates. Their approach uses brute-force matching when the input

fingerprints are small and a few minutiae are presented, and uses secondary

feature matching otherwise. Since this approach is based on minutiae, it is very

likely to produce false matches when there are very few minutiae, and it is not

applicable when there are no minutiae on the fingerprint fragments.

2.8. CHAPTER SUMMARY

From the review, it is understood that the fingerprint recognition

technique is a dominant technology in the biometric market. In relation to the

research study, the review study also reveals the fact that even though several

success stories are published with relation to complete fingerprint recognition,

a perfect partial fingerprint recognition system is not yet eminent and still

requires careful examination in terms of accuracy, speed, memory utilization

and cost.

The main complaint against the use of level 3 features (pores) is its need

for high quality and high resolution images along with improvement in terms of

accuracy. But, now, due to the advancement of high speed hardware and

sophisticated software, this complexity has been reduced to a great extent. To

improve the accuracy of the partial fingerprint systems, the research work

proposes the use of multiple non-minutiae based features, combined with level

3 pore features. The methodology used in the proposed work is presented in the

next chapter.