hand palmprint authentication system - scbaghdad.edu.iq science/ms.c/2004/hand... · ministry of...

Ministry Of Higher Education and Scientific Research Baghdad University College Of Science Computer Science Department

Hand Palmprint Authentication System

A Dissertation Submitted to the College of Science Baghdad University in the partial fulfillment of the requirement for the degree of

Master of Science in Computer science

BY

Maha MohammedAli Sharif Al-Turaihi (B.Sc. 2002)

University of Baghdad

SUPERVICED BY Dr. Loay Edwar George

(Senior Researcher)

December Dhulqa'da

2004 1425

Supervisor Certification

We certify that this dissertation was prepared under our supervision

at the Department of Computer Science of the University of Baghdad as a

partial fulfillment of the requirement needed of the degree of Master of

Science in Computer Science.

Signature:

Name: Dr. Loay Edwar George

(Senior Researcher)

(Supervisor)

Date: / /2005

Certification of the Head of the Department

In view of the available recommendation I forward this dissertation

for the debate by the examining committee.

Signature:

Name: Makia K. Hamed

(Assistance Professor)

Date: / /2005

Head of the Computer Science Department

Examining Committee Certification

We certify that we have read this dissertation and as an examining

committee, examined the student in its content and what is related to it,

and that in our opinion it meets the standard of a dissertation for the

degree of Master of Science in Computer Science.

Approved by the Council of the College of Science

Signature:

Name: Prof. A. M. Taleb

The Dean of College of Science / University of Baghdad

Date: / /2005

Signature:

Name: Dr. Loay Edwar George

(Senior Researcher)

(Supervisor)

Date: / /2005

Signature:

Name: Makia K. Hamed


(Member)

Date: / /2005

Signature:

Name: Dr. Lamia H. Khalid


(Chairman)

Date: / /2005

بسم هللا الرحمن الرحيم

(( وال يحيطون بشيء من علمه اال بما شاء وسع كرسيه السموات واالرض))

صدق هللا العظيم

سورة البقرة

قال رسول هللا محمد (صلى هللا عليه و آله و سلم):

((لو يعلم الناس ما في طلب العلم لطلبوه ولو بسفك المهج وخوض اللجج ))

صدق رسول هللا

قال اإلمام علي (عليه السالم) عن الرسول محمد (صلى هللا عليه وآله وسلم ):

((إذا مات مؤمن وترك ورقة واحدة عليها علم تكون تلك الورقة سترا بينه و بين النار وأعطاه هللا لكل حرف عليها مدينة أوسع من الدنيا سبع مرات)).

ABSTRACT Biometrics is defined as measurable physiological and/or

behavioral characteristics to verify the identity of an individual. It is the technology using image processing and pattern recognition in the security field. Recently biometric technologies are rapidly evolving and becoming security options to verify the true identity of an individual in the wide areas of businesses and organizations.

The process of identifying an individual usually based on a username and password. In security systems, authentication is distinct from authorization, which is the process of giving individuals access to system objects based on their identity. Authentication merely ensures that the individual is who he or she claims to be, but says nothing about the access rights of the individual.

Palmprint authentication is one of the relatively new physiological biometric technologies which exploit the unique features on the human palmprint, namely principal lines, wrinkles, ridges. The rich texture information of palmprint offers the effective means in person authentication due to its non-intrusive, user friendly, stable, low-resolution imaging and low cost requirements.

In the palmprint acquisition stage, users placed their palms on the platform of the scanner rather than of using a charge-coupled device (CCD). Therefore, palmprint images captured in the image acquisition stage may have variable size and orientation and also subject to noise. Moreover, the non-interesting regions (e.g. fingers, wrist, image background, etc) may affect the accurate processing and degrade the

verification performance. Therefore, image preprocessing is a crucial and necessary part before feature extraction.

This research work introduces an experimental evaluation of the effectiveness of utilizing Moment invariant and Complex moment in the application of palmprint verification.

The work was performed on database (including 25 images for 5 persons). Experimental results indicated that the performance of the system depends on the moment type. The non centralized moment invariant has the best performance among the other moments.

Contents Pages Chapter one General Introduction for Biometric

1.1 User Authentication System 1

1.2 Personal Authentication Using Biometrics 2

1.3 Identification and verification 2

1.4 Biometric and Biometrics 4

1.5 Physiological and behavioral Biometrics

1.6 Characteristics of Successful Biometric Identification Methods

5

1.7 Issues Related With Biometric System 6

1.7.1 Acceptance 7

1.7.2 Throughput 8

1.8 Biometric Applications 9

1.9 How biometric System Work? 11

1.10 Types of Biometric Systems 12

1.10.1 Physiological Biometric 12

1.10.2 Behavioral Biometric 17

1.11 Palmprint 20

1.11.1 Palmprint Features 22 1.11.2 Online and Offline Palm print System 23 1.12 Related Work 24 1.13 Aim of the work 30 1.14 Work layout 30

Chapter two Image Processing and Pattern Recognition Principles

2.1 Image Representation 31

2.1.1 Binary Image 31

2.1.2 Gray-Scale Image 32

2.1.3 Color Image 32

2.2 Digital Image file format 33

2.3 Image Smoothing 36

2.3.1 Noise Removal Using Spatial Filters 37

2.3.2 Mean Filter 37

2.4 Edge Detection 38

2.4.1 Sobel operator 38

2.5 Image Segmentation 39

2.5.1 Histogram 39

2.5.2 Thresholding 40

2.5.3 Histogram Thresholding 41

2.6 Brightness-Contrast Enhancement 44

2.6.1 Histogram Stretching 44

2.7 Chain Coding 46

2.8 Pattern Recognition 47

2.8.1 The Pattern Recognition Components 47

2.8.2 Distance Measurements 49

2.9 Pattern Recognition in Image Processing 49

2.9.1 Scene analysis recognition 50

2.9.2 2D Recognition by Moment 52

2.9.2.1 Pattern Recognition by Moment Invariants 52

2.9.2.2 Moment in General 53

2.9.2.3 Moment Invariants 54

2.10 Complex Moment 58

Chapter three Hand Palmprint Authentication System

3.1 System Description 59 3.2 Image acquisition 61 3.3 Palmprint Image Extraction Preprocessing 63 3.3.1 Gray Image 63 3.3.2 Binary Image 63

3.3.3 Smooth the Binary Image 66 3.3.4 Image Boundary 67 3.3.5 Chain Coding 69 3.3.6 Palmprint Area detection System 71 3.3.6.1 Detection of Finger's Join points and End points 71 3.3.6.1.1 Detection of Finger's Join points 71 3.3.6.1.2 Detection of Finger's End points 75 3.3.6.2 Get the Palm Image Area 78 3.3.6.3 Shifting the coordinate system 80 3.3.7 Palmprint Image Enhancement 81 3.3.8 Edge Detection Using Sobel Operator 82 3.3.9 Get Central Palmprint 84 3.4 Feature Extraction 85 3.4.1 Feature Extraction Using Moments Invariant 85 3.4.2 Feature Extraction Using Complex Moments 89 3.5 Enrollment Phase 90 3.5.1 Data base file 90 3.5.2 Mean File 91 3.5.3 Threshold File 92 3.6 Verification Phase 93

Chapter four Experimental Results and Discussion

4.1 Introduction 96 4.2 Enrollment Phase 96 4.3 Verification Phase 103

Chapter Five Conclusions and Future Work Suggestions

5.1 Discussion and Conclusions 5.2 Recommendations

References

Chapter one General Introduction for Biometrics

۱

Chapter One General Introduction for Biometrics

1.1 User Authentication System

The first line of defense a computer system has against intruder is user authentication. The user authentication system attempts to prevent unauthorized use by requiring users to validate their authorization to access the system. The validation is often accomplished with the use of a password that must be presented to the system. Other authentication system schemes require the user to present a physical key or take advantage of the uniqueness of a user's physical attributes such as fingerprints.

Classical user authentication systems have been based in something that you have (like a key, an identification card, etc.) and/or something that you know (like a password, or a PIN personal identification numbers).

With biometrics, a new user authentication paradigm is added: something that you are (e.g. fingerprints or face) or something that you do or produce (e.g., handwritten signature or voice).

Biometric recognition, as a means of personal authentication, is an emerging signal processing area focused on increasing security and convenience of use in applications where users need to be securely identified. Biometric characteristics are inherently associated with a particular individual, making them insusceptible to being forgotten or lost.

There are many different biometric traits that can be used, each with various benefits and drawbacks, depending on the application scenarios and required accuracy [1].

Biometric authentication requires comparing an enrolled biometric sample against a newly acquired biometric sample. The biometric system authenticates a person’s claimed identity from their previously enrolled


۲

pattern. It involves confirming or denying a person’s identity. An identity is claimed by inputting PIN or presentation of a token and then presenting a live sample for comparison, resulting in a match or no match according to predefined parameters. This is called “one-to-one” matching. In the identification, biometric system identifies a person from the entire enrolled population by searching a database for a match based solely on the biometrics. This is called “one-to-many" matching [12].

1.2 Personal Authentication Using Biometrics

The process of automatically associating an identity with an individual by means of some inherent personal characteristic is called biometric recognition. Traditionally, person authentication has been accomplished by associating to the person’s identity something that he/she possesses (e.g., a key, a card, etc.) or knows (e.g., a password, a PIN).

Biometric recognition adds a new dimension by associating a person’s identity with something that he/she is (or produces). Something that a person is indicates a physiological characteristic inherently associated with the person, while something that a person produces indicates trained act or skill that the person unconsciously does as a behavioral pattern [1].

1.3 Identification and Verification

In the biometrics industry, a distinction is made among the terms identification and verification. Identification and verification are, essentially synonymous terms. In both processes, a sample is presented to the biometric system during enrollment [2].

Biometric recognition is a generic term that encompasses the two main modes in which biometric systems operate:

1. Biometric identification is the task of associating a test biometric sample with one of N patterns or models that are available from a set of


۳

known or registered individuals. It is also known as the one-to-many (specifically, one-to-N) task and the output of this operation mode is normally a sorted list of candidate models, based on their degree of match with respect to the test sample [1, 4].

Identification is a one-to-many comparison in which the system attempts to find out who the sample belongs to, by comparing the sample with a database of samples in the hope of finding a match [2].

2. Biometric verification is the task of authenticating that a test biometric sample matches the pattern or model of a specific user. It is also known as the one-to-one problem, and the output is a binary decision (accept or reject), that is usually based on comparison of the match score between the test sample and the claimed user’s model or pattern to a decision threshold [1,4].

Verification is a one-to-one comparison in which the biometric system attempts to verify an individual's identity. In this case, a new biometric sample is captured and compared with the previously stored template. If the two samples match, the biometric system confirms that the applicant is who he/she claims to be [2].

The key distinction between these two approaches centers on the questions asked by the biometric system and how these fit within a given application. During identification, the biometric system asks, "Who is this?" and establishes whether a biometric record exists, and, if so, the identity of the enrollee whose sample was matched. During verification, the biometric system asks, "Is this person who he/she claims to be?" and attempts to verify the identity of someone who is using, say, a password or smart card [2].


٤

1.4 Biometric and Biometrics A biometric is a measurable, physical characteristic or personal

behavioural trait that is used to recognize the identity or authenticate the claimed identity of an enrolled user [4, 6].

The beauty of a biometric trait is that it is as unique as the individual from whom it was created, unlike a password or PIN; a biometric trait cannot be lost, stolen, or recreated. This makes biometrics an obvious antidote to identity theft, a problem that is mushrooming alongside databases of personal information [6, 8].

Our fingerprints are unique; perhaps less known is the fact that our bodies are unique in several other measurable areas as well [7].

Biometrics is the science of automatic identification or identity verification of individuals using physiological or behavioral characteristics. In computer security; biometrics refers to authentication techniques that rely on measurable physical characteristics that can be automatically checked [4, 7, 9].

The goal of any access control system is to let authorized people, not just their credentials, into specific places. Only with the use of a biometric device can this goal be achieved. A card-based access system will control the access of authorized pieces of plastic, but not who is in possession of the card. Systems using PINs require an individual only know a specific number to gain entry. But who actually entered the code cannot be determined. On the contrary, biometric devices verify who a person is by what they are, whether by hand, eye, fingerprint or voice characteristic [5].

A biometrics-based identification system should be characterized by its easy operation, short response time, and high accuracy [13].


٥

1.5 Physiological and Behavioral Biometrics

Biometric techniques can generally be classified into two main categories:

1. Physiological Biometrics Recognition based upon physical characteristics. Among the physical features measured are face, fingerprints, palm prints, hand or palmar geometry, retina and iris scanning, vein, and DNA analysis.

2. Behavioral Biometric: Not a physical characteristic. Behavioral biometrics is traits that are learned or acquired over time as differentiated from physical characteristics. Behavioral features are signature or handwriting, keystroke patterns, gait motion, gesture and voice in behavioral aspect [4, 9, 12].

A physiological characteristic inherently associated with the person, while something that a person produces indicates a trained act or skill that the person unconsciously does as a behavioral pattern [1].

1.6 Characteristics of Successful Biometric Identification Methods

The following factors are needed to have a successful biometric identification method: [11]

1. The physical characteristic should not change over the course of the person's lifetime.

2. The physical characteristic must identify the individual person uniquely.

3. The physical characteristic needs to be easily scanned or read in the field, preferably with inexpensive equipment, with an immediate result.


٦

4. The data must be easily checked against the actual person in a simple automated way.

Other characteristics that may be helpful in creating a successful biometric identification scheme are:

1) Ease of use by individuals and system operators.

2) The willing (or knowing) participation of the subject is not required.

3) Uses legacy data (such as face recognition or voice analysis).

In the development of biometric identification systems, some physical and behavioural features for recognition are required which: [6]

a) are as unique as possible, that is, an identical trait won't appear in two people: Uniqueness.

b) Occur in as many people as possible: Universality.

c) Don’t change over time: Permanence.

d) are measurable with simple technical instruments: Measurability.

e) are easy and comfortable to measure: User friendliness.

While a good biometric system should be reliable, low cost, user friendly, and require small amounts of data, no single biometric technique has yet met all of these prerequisites [10].

1.7 Issues Related With Biometric System

The body offers uniquely recognizable features in the following areas: fingerprints, voice, eyes, hands, and face. Different vendors are developing products around each of these features, and the jury is very much out on which technology is best. Success is measured according to a number of criteria, and each technology has both strengths and weaknesses.


۷

1.7.1 Acceptance The most critical factor in the success of a biometric system is user

acceptance of the device. There are several factors that impact acceptance. Among these factors are the following:

1. The device must cause no discomfort or concern for the user. This may be a subjective issue, but it is important to fully explain any concerns users may have. If people are afraid to use the device, they most likely will not use it properly, which may result in them not being granted access.

2. It must be easy to enroll people. Many get frustrated if they have to go through the process over and over again. From the start they are predisposed to reject the system.

3. The biometric device must be easy to use. People like things that are simple and intuitive. How many times have you been frustrated at a card reader that gives no indication of which way to swipe the card?

4. The biometric device must work correctly. If working properly, it does two things: keeps bad guys out and lets good guys in. Yet, no device is perfect. In the biometric world, the two errors a unit can make are letting a bad guy in and keeping a good guy out. [5]

The most important criteria are concerned with accuracy. The level of accuracy in biometric systems involves both the false-acceptance rate and the false-rejection rate.

The false-acceptance rate is the percentage of users who are not authorized in the system but who are nevertheless given access. The false-rejection rate is the opposite, the percentage of authorized users who are denied access.

These rates are useful, and biometric product vendors often cite them in their product descriptions. But they don't present a complete picture. The fact is people's physical traits change over time, especially with alterations due to


۸

accident or aging. And even in the short term, problems can occur because of humidity in the air, dirt and sweat on the user (especially with finger or hand systems), and inconsistent ways of interfacing with the system, such as not taking enough time for the system to make an accurate identification. And users of biometric systems, like the users of all systems, must be trained to use them most efficiently [7].

1.7.2 Throughput A logistical issue that should be considered when using a biometric is

the throughput, the total time it takes for a person to use the device. It is difficult for manufacturers to specify a throughput since it is application dependent. Most manufacturers specify the verification time for the reader, but that is only part of the equation.

When a person uses a biometric reader, they typically enter an ID number on a keypad. The reader prompts them to position his hand, finger or eye where the device can scan physical details. The elapsed time from presentation to identity verification is the “verification time.” Most biometric readers verify ID in less than two seconds.

However, when considering the use of biometrics for access control, one must look beyond the verification time and consider the total time it takes a person to use the reader. This includes the time it takes to enter the ID number, if required, and the time necessary to be in position to be scanned. If ID numbers must be entered, they should be kept as short as possible. If a long ID number must be used, some devices can obtain the number by reading a card that contains the ID number in the card code.

The total time required for a person to use the reader will vary between biometric devices depending on their ease of use and verification time. A card-based access system may appear faster. However, as one hand geometry user points out, “The speed difference between a card and the hand reader is


۹

about two seconds, but you make up for it (the difference) since your hand is right in front of you, verses fumbling around looking for your card.”[5].

Biometrics success is also judged via a number of other factors. Vulnerability to fraud, also known as barrier to attack, reflects how likely it is that a person can fraudulently get past the security. Long-term stability deals with issues such as whether a system is useful for very infrequent users, as well as whether or not users' characteristics alter over time. Other effectiveness measures can include factors that might interfere with the system, ease of use, and the size of the system and the amount of disk space its data takes up [7].

1.8 Biometric Applications

The ability to verify identity has become increasingly important in many areas of modern life, such as electronic government, medical administration systems, access control systems for secure areas, passenger ticketing, home office and home study environments [3].

Access control requires the ability to identify the person plus unlock a door, grant or deny access based on time restrictions, and monitor door alarms. There are a variety of ways biometrics accomplishes this task.

1. Standalone Systems: Many biometric devices are available in a standalone configuration. Such devices are not only a biometric, but also a complete controller for a single door. Users are enrolled at the unit and their biometric template is stored locally for subsequent comparison. The actual comparison is accomplished within the unit and a lock output is energized depending on the outcome.

2. Networked Systems: Many access control applications have a need to manage more than one door. While multiple stand alone units could be employed, a network of biometric readers is much more feasible. By networking the systems together and then connecting them to a


۱۰

computer, several advantages are available to users. The most obvious is centralized monitoring of the system. Alarm conditions and activity for all the doors in the system are reported back to the PC. All transactions are stored on the computer’s disk drive and can be recalled for a variety of user-customized reports.

Networked systems also provide convenient template management. Although a user enrolls at one location, his template is available at other authorized locations. Deletion of a user or changes in his access profile is simply entered at the PC. Some biometric systems store all information in the PC where template comparisons are also performed. Others distribute template information to the individual readers at each door. Either way, the net effect of template management is the same.

3. Smart Card Systems: Smart cards raise the bar even higher, providing additional capabilities and flexibilities. As costs begin to come down and usage is more widespread, biometric devices can leverage their secure data storage.

For example, a smart card can store both the user’s ID number and hand geometry template on the card. Because of this, there is no need to distribute hand templates across a network of hand readers or require the access control system to manage biometric templates. This means integration to any existing access control application is greatly simplified and additional network infrastructure costs are eliminated. Since the template only resides on the card, the solution also eases individual privacy concerns.

Providing the best of smart cards and biometrics, the solution provides dual authentication by requesting both the right card and the right person. A smart card reader is embedded into the biometric reader. A plastic cardholder is affixed to the side of the unit. The verification process takes approximately one second and is virtually foolproof [5].


۱۱

1.9 How Biometric Systems Work?

Biometric systems consist of both hardware and software; the hardware captures the salient human characteristic, and the software interprets the resulting data and determines acceptability. The crucial step in building an effective biometric system is enrollment. During enrollment each user, beginning with the administrator who controls the system, provides samples of that system's specific biometric characteristic by interacting with the scanning hardware.

As an example, you touch a finger to a fingerprint reader or stand in front of a camera so it can capture face or eye features. The system then extracts the appropriate features from the scan and stores the data as a template.

You then interact with the biometric device again, and the system verifies that the data corresponds to the template. If the software fails to get a match, more tries may be needed, just as dictation software learns to recognize the user's speech patterns over time. Once this procedure is complete, the system is operational.

The next time you try to access the system, you are scanned by whatever device is being used (you might be asked to supply a user name as well), and the hardware passes the data to the software, which checks the user templates. If there is a match, you are granted access; otherwise, a message reports that the system can't identify the user. If access is granted, it is based on your profile. If you are trying to log on to a Windows 98 or NT machine, for example, the system will unlock just as if you had typed your user name and password at the log-in prompt [7].


۱۲

1.10 Types of Biometric Systems The utilized biometric systems could be categorized in to the following

types: 1.10.1 Physiological Biometrics

This type of biometric involves the following: 1. Fingerprint Biometrics

Fingerprint matching is one of the most widespread biometric solutions and is based on the ridge structure of the epidermis of each fingertip and its peculiar distribution of minutiae points, which is preserved almost unalterable all through a person’s life. Traditional ink-and-roll fingerprint images have been replaced by electronic scanning through different technologies (optical, capacitive, ultrasonic, etc.). Unfortunately, these devices may introduce some degree of variability on fingerprint patterns, due to several effects, like the sensitivity to finger temperature or sweat, the distortion due to pressing on the planar surface, fingertip placement, or the size-limited nature of the device that adds position variability to the fingerprint. Moreover, there are some population groups with special problems, like some manual workers that make intensive use of manual tools that can lead to fingerprint damaging by friction and erosion, an issue that can affect up to 5% of the population.

Features used to represent fingerprints for person authentication purposes are predominantly minutiae based. These are typically end point or bifurcation ridges, the relative position of these minutiae points constitute a personal trait [1].

This system consists of a hardware scanner and recognition software records specific fingerprint characteristics, saves each user's data in a template, and then refers to the templates when the user next tries to gain access.

The fingerprint scanners shine a light through a prism that reflects off your finger to a charge-coupled device (CCD), creating an image that gets processed by an onboard computer. It's important to note that the actual


۱۳

fingerprint image is not recorded. Instead, the devices perform a reduction of the image to data points, called minutiae, which describe the fingerprint layout, called a template.

It works by matching relationships between minutiae the points on your fingertips where print ridges end or divide. Scanners come embedded in keyboards and mice, so setting up the system is easy and inexpensive (keyboards cost about $150). And it's already used in law enforcement.

Some advantages of fingerprint identification are:

1. High Accuracy Fingerprint technology is accurate; however, a fingerprint scanner might deny access to your PC if you place too much or too little pressure on it.

2. Easy Installation Fingerprint scanners are as simple to set up as a new keyboard and require no training to use [7].

Results of the 2000 and 2002 Fingerprint Verification Competitions (FVC2000 and FVC2002, respectively) reveal that even with medium- to high quality images, only a few technologies show good performance.

This means there is stillroom for algorithmic improvement in terms of image processing to extract the salient features and match them despite intra-class variability.

An improvement is also needed regarding acquisition devices, which provide size reduced, rather poor-quality images [1].

Due to throughput concerns, fingerprint access control may be best applied in smaller user populations. Because of cost and size, they are a perfect choice for single person verification applications, such as in logical access control, where they are used to log onto PCs or computer networks. This is certainly a fast growing application for this technology [5].


۱٤

2. Face Biometrics The face of a person is considered to be the most immediate and

transparent biometric modality for physical authentication applications. As far as adequate camera positioning can mitigate issues with users properly looking into the camera, cooperativeness is not in some cases a crucial issue.

Nevertheless, in other uncontrolled situations, like when trying to detect a particular face in a crowd, cooperativeness is an important issue. This makes face recognition a highly desirable biometric modality, and extraordinary research efforts have been undertaken in the last decade. A wide choice of techniques has been proposed to meet the demands of automatic person authentication by their faces.

Significant motivations for its use include the following:

1. Face is a modality that humans largely depend on to authenticate other humans; consequently, every human is a putative expert in face recognition from infancy.

2. Face is a modality that requires no or only weak cooperation to be useful.

3. Face authentication can be advantageously included in multimodal systems, not only for authentication purposes but also to confirm the liveness of the signal source of fingerprints, voice, etc [1].

Face recognition can detect a person using static digital photographs or live video feeds. The system locates and tracks a persons head, the main features used for description of faces are either geometric metrics of the face, like distances between the nose and mouth, the shape of the face determined by distances between the eyes, ears and nose and other facial characteristics are put into a template.


۱٥

In summary, the system captures your face in moving video, isolates features, and identifies them on the fly, and the system is smart enough to recognize your face even if you forgot to shave or your eyes are bloodshot [4, 5, 7].

3. Iris Recognition

The pattern of the iris, the band of tissue that surrounds the pupil of the eye, is complex, with a variety of characteristics unique in each person.

An iris recognition system uses a video camera to capture the sample and software to compare the resulting data against stored templates [7].

Left and right iris patterns of a given person are different. Also, iris patterns between identical twins are different, iris patterns are quite unique and do not change with age. Unlike face scan technology, which can leverage existing photo- or video-camera technology, iris scan deployments require specialized devices including, in some cases, infrared illumination and may be perceived as invasive by users who are required to be very collaborative [1].

4. Retina Recognition

Probably the single most secure of all, these biometric systems work with the retina, the layer of blood vessels located at the back of the eye.

The retinal image is difficult to capture, and during enrollment the user must focus on a point while holding very still so the camera can perform the capture properly. The only thing that is actually determined is the pattern of the blood vessels, but since this pattern is unique in each person, identification can be precise.

The two eye-based systems, iris and retina, are generally considered to offer the best security, because of the distinctiveness of the patterns and the quality of the capture devices [7].


۱٦

5. Palmprint Recognition Palmprints are stable and show high accuracy in representing each

individual’s identity. Thus, they have been commonly used in law enforcement and forensic environments. Since the surface of the palmprint is larger than the fingerprint itself, a higher quantity of identifying features can be extracted from the palmprint. Moreover, users consider hand biometrics as being user friendly, easy to use, and convenient.

Palmprint acquisition is based on standard charge-coupled device (CCD) based optical scanning. Although some acquisition procedures imply pressing on a glass panel (inducing an elastic distortion on the palmprint), some others do not. Those who do not however, must solve the liveness issue separately.

Palmprint features can be divided into three different categories: (a) point features, which include minutiae features from ridges existing in the palm, and delta point features, from delta regions found in the finger-root region; (b) line features, which include the three relevant palmprint principal lines, due to flexing the hand and wrist in the palm, and other wrinkle lines and curves (thin and irregular); and, (c) texture features of the skin [1].

6. Hand Geometry

Hand geometry recognition is based on the extraction of a hand pattern that incorporates parameters like finger length, width, thickness, curvatures, or relative location. To obtain these features, an image of the silhouetted hand is needed. The process of capturing this information is normally accomplished through CCD cameras and infrared illumination; the user puts his/her hand on a highly reflective surface, such as a platen, performing an orthographic scanning, consisting of top and side views of the hand shape. Surface details like texture and fingerprints are ignored for this purpose. Specific hand positioning is forced by using inter-finger pegs or locator pins.


۱۷

Hand geometry requires high collaboration from the users, as the hand must be kept flat while scanning. Fingernails (e.g., on females) that can potentially deteriorate intra-class variability must also be coped with by the system [1].

7. Finger Geometry

These devices are similar to hand geometry systems. The user places one or two fingers beneath a camera that captures the shapes and lengths of the areas of the finger and the knuckles. The system captures a three-dimensional image and matches the data against the stored templates to determine identity [7].

The leading finger geometry system uses a special camera, which creates a “Profile Photo” using a virtual 3-dimensional image of any 2 fingers. It does not record a fingerprint. The system electronically captures and processes a digital profile of the 2 fingers. The enrollment data (template) of only 20 bytes may be stored inside the camera module, can be transmitted to other units and/or data stores [4].

1.10.2 Behavioral Biometrics

This type of biometric implies the following: 1. Voice Biometrics

The speech signal conveys many levels of information to the listener. At the primary level, speech conveys a message via words, but at other levels speech conveys information about the language being spoken and the emotion, gender, and, generally, the identity of the speaker. While speech recognition aims at recognizing the words spoken in speech, the goal of automatic speaker recognition systems is to extract, characterize, and recognize the information in the speech signal conveying speaker identity.


۱۸

Depending on the level of user cooperation and control in an application, the speech used for these tasks can be either text dependent or text independent.

In a text-dependent application, the recognition system has prior knowledge of the text to be spoken and it is expected that the user will cooperatively speak this text.

In a text-independent application, there is no prior knowledge by the system of the text to be spoken, such as when using extemporaneous speech. Text-independent recognition is more difficult but also more flexible, for example, allowing verification of a speaker while he/she is conducting other speech interactions (background verification) [1].

This method captures the sound of the speaker's voice as well as the linguistic behaviors. Its primary use is in telephone-based security applications, but its accuracy can be affected by such things as extraneous noises and the effects of illness or fatigue on the voice.

One obvious problem with voice recognition is fraud; the system can be fooled by a tape of someone's voice. For this reason, advanced voice systems can extend the verification process by giving the user longer and more difficult phrases to read aloud, or requesting a different phrase to be read each time. This does increase the time needed for verification, however, and thus cuts into the system's overall usability.

Voice authenticators use a telephone or microphone to record a user's voice pattern, and then use that pattern to validate the person. Since these software systems rely on very low-cost devices, they are generally the least expensive systems to implement for large numbers of users [4].


۱۹

2. Signature Recognition

Signature verification systems have one major thing going for them: public acceptance. On everything from the Declaration of Independence to a credit card slip, people tend to accept a person's signature as proof of identity.

Actually, signature recognition systems, also called dynamic signature verification systems, go far beyond simply looking at the shape of a signature. They measure both the distinguishing features of the signature and the distinguishing features of the process of signing.

These features include pen pressure, speed, and the points at which the pen is lifted from the paper. These behavioral patterns are captured through a specially designed pen or tablet (or both) and compared with a template of process patterns.

The problem is that our signatures vary significantly over time and from one instance to another, so strong accuracy requires multiple samples and an extended verification process [4].

3. Keyboard Dynamics

Keyboard dynamics technology measures dwell time, the length of time you hold down each key, as well as flight time, the time it takes you to move between keys. Taken over the course of several login sessions, these two metrics produce a measurement of rhythm unique to each user.

These are the major biometric systems currently under development, but they're not the only ones. Researchers are developing or examining the feasibility of systems based on the analysis of DNA, vein patterns, and even bodily odors [4].


۲۰

1.11 Palmprint There are a number of distinguishable features of the human hand.

These features are: the size and form of the lines on the palm; the lengths of the fingers in proportion to the palm; the length and mobility of the thumb; the hollowness of the palm; the spread of the digits; and the general size and shape of the hand [16].

In particular, the palmprint is one of the most reliable physiological characteristics that can be used to distinguish between individuals. It’s unique to every individual; even our own two hands are never quite alike

Palmprint-based systems are more user friendly, more cost effective, and require fewer data signatures than traditional fingerprint-based identification systems.

Fingerprint identification is the most well-known and widespread biometric method. It is very reliable, but fingerprint-capturing devices are expensive and the stored data is large. Furthermore, it can be difficult to extract some minutia features from some hands, for example from the hands of manual laborers and elderly people, whose fingers are heavily worn down.

In some applications, several other methods may be better than fingerprints, since they require fewer data signatures, are less expensive and less intrusive, and avoid the stigma of fingerprinting, which makes people feel like criminals.

In recent years, iris-based verification has been successfully developed, but it suffers from the discomfort of iris picture capturing that requires users to put their eyes before a camera. Thus, there is a demand for a new automatic personal identification system.

Similar to fingerprint and iris verification, the palmprint is one of the most reliable means in personal identification because of its stability and uniqueness.


۲۱

In fact, palmprints have been taken as a human identifier for a long time. Using hand features as a base for identity verification is relatively user-friendly and convenient.

Some hand-geometry- based identification systems have also been developed but the hand geometry features are not unique and the systems are less accurate in a large database [10].

Table (1.1) compares palmprints with both fingerprints and hand geometry in terms of the accuracy and feature complexity for personal identification.

The reasons to choose hand features as a base for identity verification is originated by its user friendliness, environment flexibility, and discriminating ability.

Palmprints-based identifiers are not only user friendly; they use less data amount and can be operated using cheap electronic imaging device [13].

Public Acceptance Accuracy Level Features

Fingerprint Medium High Minutiae Points

Hand High Low 3-D Geometry

Palmprint High High Line & texture

Table (1.1) Palmprint versus fingerprints and hand geometry


۲۲

1.11.1 Palm Features There are a number of attributes, which make the lines on a person’s

hand distinguishable from the lines on another. These include: color, clarity, and length, position within the palm, continuity and variations in thickness [16].

The palm is the inner surface of the hand between the wrist and the fingers. As illustrated in the figure (1.1), the main patterns in a palmprint can be generalized as principal lines, wrinkles and creases.

There are usually three principal lines in a palmprint: the heart line, the headline, and the lifeline. These lines vary little over time, and their shapes and locations on the palm are the most important physiological features for individual identification. Wrinkles are much thinner than the principal lines and much more irregular. Creases are the relatively detailed features that exist all over the palmprint, just like the ridges do in a fingerprint.

Generally speaking, for identification tasks the features of principal lines and wrinkles can be exploited and derived from a low-resolution palmprint image. Although some crease-based palmprint recognition methods

Figure (1.1) the main patterns in a palmprint


۲۳

have been proposed, they require rather fine resolution imaging and consume a large amount of data. Thus, for personal identification, creases are not as robust and persistent as principal lines.

Principal lines, wrinkles, ridges, minutiae points, singular points, and texture are regarded as useful features for palmprint representation. Various features can be extracted at different image resolutions. For features such as minutiae points, ridges, and singular points, a high-resolution image, with at least 400 dpi (dots per inch), is required for feature extraction However, features like principal lines and wrinkles can be obtained from a low-resolution palmprint image with less than 100 dpi.

For civil and commercial applications, low-resolution palmprint images are more suitable than high-resolution images because of their smaller file sizes, which result in shorter computation times during preprocessing and feature extraction. Therefore, they are useful for many real-time palmprint applications [15].

1.11.2 Online and Off line Palmprint Systems

Automatic palmprint identification systems can be classified into two categories: online and offline.

An online system captures palmprint images using a palmprint capture sensor that is directly connected to a computer for real-time processing.

An offline palmprint identification system usually processes previously captured palmprint images, which are often obtained from inked palmprints that were digitized by a digital scanner. In the past few years, some researchers have worked on offline palmprint images and have obtained useful results.

Recently, several researchers have started to work on online palmprint images that are captured by CCD (charge coupled device) cameras or digital scanners [14, 15].


۲٤

1.12 Related Work In 1997 Boles and Chu, they presented a prototype system for human

identification using palm images. Hough transform was used to detect palm features as approximated straight lines, circles and curves in 2D images. Translation and rotation invariance were achieved by using the edge of the palm as a reference for all feature measurements. An image capture platform was constructed such that variation of the camera to palm distance is minimized in order to eliminate the need to compensate for scaling transformation.

The system was tested by taking 30 images from three people (10 images each) and running comparisons between them. The 10 images of the same user were tested against each other, and then five images from this user were tested against 5 images from each of the other two users. This brings the total number of tests to sixty. It was found that the lines of the palm provided a good basis for identification.

The system correctly identified a user 77% of the time and correctly rejected a user 90% of the time. These results suggested that the current system cannot be used as the only source of identification. The proposed system was modified to improve its accuracy and robustness, the results obtained under improved lighting conditions [16].

In 2002 Kong and Zhnag they developed a new feature extraction method based on low-resolution palmprint images. A 2-D Gabor filter was used to obtain the texture information. Two palmprint images are compared in term of their hamming distance. Recently, a CCD camera-based palmprint capture device has been developed in Biometric Research center, Hong Kong palmprint image captured by their self-designed capturing device, which a low resolution technique (65 dpi) is adopted to reduce an image size.

The database for testing their method contains 425 images from 95 persons. The sub images of the palmprint images is 64 by 64.


۲٥

The performance of the proposed method under different thresholds which control the false accept rate and false reject rate. For the optimal cases in their database, the false reject rate is 0.91% with 0% false accepts rate [14].

In November 2002 Wu, Wang and Zhang, They utilized the fact that principle lines, wrinkles and ridges have different resolutions motivated them to analyze the palmprint using multi-resolution analysis method. A novel palmprint feature, named wavelet energy features, is defined employing wavelet, which is a powerful tool of multi-resolution analysis.

WEF (Wavelet Energy Feature) can reflect the wavelet energy distribution of the principle lines, wrinkles and ridges in several directions at different wavelet decomposition level (scale), so its ability to discriminate palms is very strong. Easiness to compute is another virtue of WEF.

A database of 1,000 palmprint images from 200 palms were used in experiments. These palmprints were taken from different ages, 5 images are captured from each palm in which four are used to train template and the remaining one is the testing sample. The resolution of these images is 320 x 240 with 256 grayscales. The 128 x 128 central part of each palmprint is croped to represent the whole one. Each detail image is divided into 4 x 4 non-overlapping blocks. The city block distance is used to describe the similarity between WEFs. All the experiments are conducted with Matlab 6.1 on PIII 1G, 256M RAM PC.

They tested their method using different wavelets and different wavelet decomposition levels. Daubechies wavelet and Symmlet orthonormal wavelet and vanishing moments of the corresponding wavelet filters used, the highest recognition rate, 99.5%, is obtained at 4 level decomposition when the wavelet is chosen as symmlet and the vanishing moment is 4.

Also 99.0% recognition rates are gotten when using Harr wavelet (3 level), Daubechies wavelet with 9 and 10 vanishing moments (4 level) and


۲٦

Symmlet wavelet with 9 vanishing moment (4-6 level). Considering other factors such as storage and computation complex, etc., Harr wavelet, the simplest wavelet, is the best choice to palmprint recognition for their database when the images composed to 3 scales.

The total training time using Harr wavelet at 3 level is 187 seconds and the average testing time is 0.29 second (not include the orientation time). That’s show that Harr wavelet decomposition to 3 scales is the most suitable for online palmprint recognition [17].

In 2003 Doi and Yamanaka introduced a practical method of reliable and real-time authentication. Finger geometry and feature extraction of the palmar flexion creases were integrated in a few numbers of discrete points for faster and robust processing.

A video image of either palm, palm placed freely facing toward a near infrared video camera in front of a low-reflective board, is acquired.

Fingers were brought together without any constraints. Discrete feature point involves intersection points of the three digital (finger) flexion creases on the four finger skeletal lines and intersection points of the major palmar flexion creases on the extended finger skeletal lines, and orientations of the creases at the points. These metrics define the feature vectors for matching.

Matching results were perfect for 50 subjects so far. This point wise processing, extracting enough feature from non contacting video image, requiring no time-consumptive palm print image analysis, and requiring less than one second processing time, will contribute to a real-time and reliable authentication. The overall time, from image acquisition to the matched result output for each subject is about 0.9 seconds using a 2.0 GHz Pentium 4 processor [12].

In September 2003 Zhang, Kong, You, and Wong, They developed an online palmprint identification system for real-time personal identification by applying a novel CCD camera-based palmprint device to capture the palmprint


۲۷

images. A preprocessing algorithm extracts a central area from a palmprint image for feature extraction.

To represent a low-resolution palmprint image and match different palmprint images, they extended the use of 2D Gabor phase coding to represent a palmprint image using its texture feature, and applied a normalized hamming distance for the matching measurement.

Using this representation, the total size of a palmprint feature and its mask was reduced to 384 bytes. In their palmprint database of 7,752 palmprint images from 386 different palms, they can achieved high genuine (98 percent) and low false acceptance (0.04 percent) verification rates, and its equal error rate is 0.6 percent, which is comparable with all other palmprint recognition approaches.

This result is also comparable with other hand-based biometrics, such as hand geometry and fingerprint verification. For 1-to-100 identification, their system can operate at a 0.1 percent false acceptance rate and a reasonable 97 percent genuine acceptance rate.

The system is implemented using visual C++ 6.0, the execution time for the whole process, including preprocessing, feature extraction, and final matching, is between 0.6 and 1.1 seconds on a Pentium III processor (500MHz) [15].

In February 2004 You Kong, Zhang and Cheung, They presented a new approach to personal identification using palmprints. To tackle the key issues such as feature extraction, representation, indexing, similarity measurement, and fast search for the best match, they proposed a hierarchical multi feature-coding scheme to facilitate coarse-to-fine matching for efficient and effective palmprint verification and identification in a large database.

In their approach, four-level features were defined: Level-1 global geometry-based key point distance, Level-2 global texture energy, Level-3 fuzzy “interest” line, and Level-4 local directional texture energy possesses a


۲۸

large variance between different classes while maintaining a high compactness within the class.

The coarse-level classification by Level-1, Level-2, and Level-3 features was effective and essential to reduce the number of samples significantly for further processing at the fine level. The Level-4 local texture leads to a fast search for the best match.

In contrast to the existing systems that employ a fixed mechanism for feature extraction and similarity measurement, they extracted multiple features and adopted different matching criteria at different levels to achieve high performance by a coarse-to-fine guided search.

The palmprint image samples used for the testing are 284384´ pixel in size with a resolution of 75 dpi and 256-level grayscale. In their palmprint image database, 7752 palmprint images from 386 different palms were stored. The images were collected on the two occasions with an average time interval 69 days.

The palmprint samples were collected from both female and male adults with an age range of 18–50. Some image samples were automatically removed during the preprocessing due to the inappropriate placement of a palm for data acquisition. The total number of images for testing is 5437 images.

The proposed system was implemented by using Visual C++6.0 on an embedded Intel Pentium III processor (500 MHz) PC, the execution time of the simulation of their hierarchical coding scheme for a large database palmprint samples was 2.8 s while the traditional sequential approach requires 6.7 s with 4.5% verification equal error rate [13].

In June 2004 Zhang and Zhang developed a Characterization of Palmprints by Wavelet Signatures via Directional Context Modeling. They presented a statistical approach to palmprint identification that uses low-resolution images, transformed the palmprints into the wavelet domain and


۲۹

then identified the predominant structures using context modeling according to the appearances of the principal lines in each sub band.

By using the interested context image, they characterized an input palmprint with a set of statistical signatures. Some of the signatures were used to classify the palmprints, and all the signatures were used to calculate the WDs between the palmprints and the database. The proposed scheme provided an adequate statistical description of the principal lines and heavy wrinkles, which convey considerable information for purposes of individual identification. The experiments were performed using two hundreds palmprint images from fifty persons.

The 50 individuals were classified into eight categories. The corrected recognition rate was as high as 98%. The proposed scheme works very well at suppressing the false identification rate and robust to the threshold. These results were encouraging and the scheme will be tested on a much larger database. The main limitation of the approach is that the used signatures are global measurements, and the signatures of some palmprints are very similar [10].


۳۰

1.13 Aim of Thesis The aim of this thesis is to design a system that uses palmprint features of a person's hand that is one of the most reliable physiological characteristic that can be used to distinguish between individuals, this system named "Hand Palmprint Authentication System".

Palmprint features are extracted from colored digital image taken when a user place his\her hand on a platform surface device, the available device used is a scanner .Then the proposed pattern recognition system has been used to match the extracted features with the corresponding features extracted from the same person (that is previously enrolled by the system and its extracted features values are stored in a database file).

1.14 Thesis Layout Chapter One: "General Introduction for Biometrics" presents an over view to biometric systems and introduce the basic concepts of palmprint recognition system.

Chapter Two: "Image processing and pattern recognition" presents the basic image processing and pattern recognition principles that are used to extract the palm features from the hand image, and to match these features with the features stored in the database.

Chapter Three: "The Design of Palmprint Authentication System" presents the designed steps of the system for both preprocessing and feature extraction stages.

Chapter Four: "The Experimental Result and Discussion" presents the implementation results of some tests applied on the proposed system, and a discussion on the obtained results is given.

Chapter Five: "Conclusions and Recommendations" presents the derived conclusions a bound the performance of the proposed system, and some recommendation for future work are given.

Chapter two Image processing and pattern recognition

۳۱

Chapter Two Image Processing and Pattern Recognition Principles

2.1 Image Representation

The digital image I(r,c) is represented as two dimensional array of data, where each pixel value corresponds to the brightness of the image at that point (r,c). In linear a algebra terms, a two-dimensional array like our image model I(r,c) is called matrix, while one row (or column) matrix is called vector.

This image model is for monochrome (one-color this is what we normally refer to as black and white) image data, but we have other types of image data require extensions or modifications to this model. The image model types we will consider are: (1) binary, (2) gray-scale, (3) color [18].

2.1.1 Binary Image

Binary images are the simplest type of images and can take on two values, typically black and white, or ‘0’and ‘1’. A binary image is referred to as a 1 bit/pixel image because it takes only 1 binary digit to represent each pixel. These types of images are most frequently used in computer vision applications where the only information required for the task is general shape, or outline information.

Binary images are often created from gray–scale images via a threshold operation, where every pixel above the threshold value is turned white (‘1’) and those below it are turned to black (‘0’) [18].

Binary images consist of groups of pixels selected on the basis of some property. The selection may be performed by thresholding brightness values, perhaps using several grey scale images containing different color bands, or processed to extract texture or other information.


۳۲

The goal of binarization is to separate features from background, so that counting, measurement, or matching operations can be performed.

It is possible to use a binary image as a mask to modify a grey scale image. This is most often done to blank out (to either white or black, but generally to background) some portion of the grey scale image, either to create a display in which only the regions of interest are visible or to select regions whose brightness, density, and so forth are to be measured [31].

2.1.2 Gray - Scale Image

Gray–scale images are referred to as monochrome, or one-color, images. They contain brightness information only, no color information. The number of bits used for each pixel determines the number of different brightness level available. The typical image contains 8 bit/pixel data, which allows us to have 256 (0-255) different brightness (gray) levels [18].

2.1.3 Color Image

Color image can be modeled as three-band monochrome image data, where each band of data corresponds to a different color. The actual information stored in the digital image data is the brightness information in each spectral band. When the image is displayed, the corresponding brightness information is displayed on the screen by picture elements that emit light energy corresponding to that particular color. Typical color image is represented as red, green, and blue, or RGB image. Using the 8-bit monochrome standard as a model, the corresponding color image would have 24 bits/ pixel – 8-bits for each of the three color bands (red, green, and blue) [18].


۳۳

2.2 Digital Image file format

Graphic images that have been processed by a computer can usually be divided into two distinct categories. Such images are either bitmap files or vector graphics.

As a general rule scanned images are bitmap files, while drawings made in applications, like Corel Draw or Illustrator, are saved as vector graphics. But you can convert images between these two data types and it is even possible to mix them in a file. This sometimes confuses people [23].

In computer graphics, types of image data are divided into two primary categories: bitmap and vector. Bitmap images (also called raster images) can be represented by our image model I(r,c), where we have pixel data and corresponding brightness values stored in some file format. Vector images refer to methods of representing lines, curves, and shapes by storing only the key points.

These key points are sufficient to define the shapes, and the process of turning these into an image is called rendering. After the image has been rendered, it can be thought of as being in bitmap format, where each pixel has specific values associated with it [18].

BMP Image Format:

BMP is a standard file format for computers running the Windows operating system. The format was developed by Microsoft for storing bitmap files in a device-independent bitmap (DIB) format that will allow Windows to display the bitmap on any type of display device. The term "device independent" means that the bitmap specifies pixel color in a form independent of the method used by a display to represent color [20].

BMP files are historic (but still commonly used) file format for the historic (but still commonly used) operating system called "Windows". BMP images can range from black and white (1 byte per pixel) up to 24 bit color (16.7 million colors) [21].


۳٤

BMP Structure: A BMP file consists of either 3 or 4 parts as shown in the figure (2.1).

The first part is a header, this is followed by information section, if the image is indexed color then the palette follows, and last of all is the pixel data. The position of the image data with respect to the start of the file is contained in the header. Information such as the image width and height, the type of compression, the number of colors are contained in the information header.

Bitmap Header:

The bitmap-file header contains information about the type, size, layout, dimensions, compression type, and color format of a device-independent bitmap file [20, 22]. The byte wise distribution of this header is illustrated in Table (2.1).

Header

Information header

Optional palette

Image data

Figure (2.1) Structure of BMP file.

14 40

4 x No. of colors


۳٥

Color Table: The color table (palette) is used to store the color components of red,

green and blue for the used color set. This color table is not present for bitmaps with 24 color bits because for such files each pixel is represented by 24-bit red-green-blue (RGB) values in the actual bitmap data area. The colors in the table should appear in order of importance. This helps a display driver render a bitmap on a device that cannot display as many colors as there are in the bitmap [22].

Bitmap Bits:

Immediately following the color table there is an array of byte values representing consecutive rows or scan lines of the bitmap. Each scan line consists of consecutive bytes representing the pixels in the scan line in left-to-right order. The number of bytes representing a scan line depends on the color format and the width (in pixels) of the bitmap.

Item Bytes signature - 'BM' 2 size of file in bytes 4 Reserved must be 0 4 offset to start image data in byte 4 size of bitmapinfoheader structure 4 width of image in pixel 4 height of image in pixel 4 number of color planes always 1 2 Bits per pixel (1,4,8 or 24) 2 compression flag (0=none,1=RLE-8 ,2=RLE-4) 4 Compressed image 4 Horizontal resolution 4 vertical resolution 4 color table size 4 important color count 4

Table (2.1) Bitwise distribution of the BMP header.


۳٦

The scan lines are stored from bottom up. This means that the first byte in the array represents the pixels in the lower-left corner of the bitmap and the last byte represents the pixels in the upper right corner [22].

24 bit Image Data:

The simplest data to read is 24 bit true color images. In this case the image data follows immediately after the information header, that is, there is no color palette. It consists of three bytes per pixel in b, g, and r order [21].

Image Resolution:

Resolution is the overall measurement of image quality. It is expressed as the number of dots or pixel per inch. Increase the dot or pixel per inch in the image leads to higher resolution and quality, but increase its file size.

2.3 Image Smoothing

Image smoothing is used for two primary purposes: to give an image a softer or special effect or to eliminate noise. Image smoothing is accomplished in the spatial domain by considering a pixel and its neighbors and eliminating any extreme values in this group. This is done by various types of mean and median filters.

In the frequency domain, image smoothing is accomplished by some form of lowpass filtering. Because the high spatial frequencies contain the detail, including edge, information, and the elimination of this information via lowpass filtering will provide a smoother image.


۳۷

2.3.1 Noise Removal Using Spatial Filters What is noise? Noise is any undesired information that contaminates

an image. Noise appears in images from a variety of sources. Spatial filters can effectively used to remove various types of noise in digital images. These spatial filters typically operate on small neighborhoods, 3x3 to 11x11, and some can be implemented as convolution masks.

The two primary categories of spatial filters for noise removal are order filters and mean filters. The order filters are implemented by arranging the neighborhood pixels in order from smallest to largest gray-level value and using this ordering to select the "median" value, while mean filters determine, in one sense or another an average value.

2.3.2 Mean Filter The mean filters function by finding some form of an average within the NxN window, using the sliding window concept to process the entire image. The most basic mean filter is the arithmetic mean filter, which finds the arithmetic average of the pixel values in the window, as follows:

å=Wcr

2 crdN1MeanArithmetic

),(),( )1.2.....(....................

Where 2N is the number of pixels in the N xN window (W)

The arithmetic mean filter smoothes out local variations within an image, so it is essentially a lowpass filter. It can be implemented with a convolution mask where all mask coefficients are 2

1N .This filter will tend

to blur an image while mitigating the noise effects [18].

1/9 1/9 1/9

1/9 1/9 1/9

1/9 1/9 1/9

Figure (2.2) The filter coefficients of the Mean (3x3) mask.


۳۸

2.4 Edge Detection The large change in image brightness over a short spatial distance

indicates the presence of edge. Some edge detection operators return orientation information (information about the direction of the edge), whereas others only return orientation information about the existence of an edge at each point. Edge detection methods are used as a first step in the line detection process.

Edge detection is also used to find complex object boundaries by mark potential edge points corresponding to places in an image when rapid changes in brightness occur. After these edge points have been marked, they can be merged to form lines and object outlines.

Edge detection operators based on the idea that edge information in the image is found by looking at the relationship a pixel has with its neighbors. If a pixel's gray level value is similar to those around it, there is probably not an edge at that point. However, if a pixel has neighbors with widely varying gray levels, it may represent an edge point. In other words, an edge is defined by a discontinuity in gray–level value. Ideally, an edge separates two distinct objects. In practice, apparent edges are caused by changes in color or texture or by specific lighting conditions present during the image acquisition process.

2.4.1 Sobel operator The Sobel edge detection masks look for the edges in both the

horizontal and vertical directions and then combine this information into a single metric. The masks are as follows:

-1 0 1

-2 0 2

-1 0 1

-1 -2 -1

0 0 0

1 2 1

ROW MASK COULMN MASK ROW MASK

Figure (2.3) Sobel operator masks.


۳۹

These masks are each convolved with the image. At each pixel location we have two numbers: s1, corresponding to result form the row mask, and s2, form the column mask. We use these numbers to compute two metrics, the edge magnitude and the edge direction, which are defined as follows:

22

21 ssudeEdgeMagnit += )2.2.........(..........

úû

ùêë

é= -

2

11

sstanionEdgeDirect )3.2.........(..........

The edge direction is perpendicular on the edge itself because the specified direction is the direction of the gradient, along which the gray levels are changing [18].

2.5 Image Segmentation

Image segmentation serves as the key of image analysis and pattern recognition. It's a process of dividing an image into different regions such that each region is homogeneous, but the union of any two regions is not. 2.5.1 Histogram

The histogram presents the frequency of grayscale values in an image. Some examples are given below.

P(f) P(f) P(f)

T Figure(2.4) Histogram


٤۰

The Histogram of an image is a plot of the gray–level values versus the number of pixel at that value. The shape of the histogram provides us with information about the nature of the image, or sub image that we are considering [19].

The gray-level histogram of an image is the distribution of the gray levels in an image. In general, a histogram with a small spread has a low contrast, and a histogram with a wide spread has high contrast, whereas an image with its histogram clustered at low end of the range is dark, and a histogram with the values clustered at the high end of the range corresponds to a bright image [18].

2.5.2 Thresholding

Thresolding is one of the most important approaches to image segmentation. The simplest method of gray-level reduction is the thresholding. We select a threshold gray level and set everything above that value equal to ‘1’ (255 for 8-bit data) and every thing at or below the threshold equal to ‘0’(also called Global thresholding).

This effectively turns a gray-level image into binary image and is often used as a preprocessing step in the extraction of object features such as shape, area, or perimeter [19].

Thresholding is a fundamental technique applied in many image processing applications. It is essentially a pixel classification problem in which the main objective is to separate the pixels of a given image into two classes, namely object and background. While one includes pixels with gray values that are below or equal to a certain threshold, the other includes those with gray values above the threshold.

In the past four decades, many threshold selection techniques have been reported in the literature, however the selection of an appropriate one can be a difficult task. The problem is that different algorithms typically produce different results since they are based on different assumptions about the image content. For instance, some require the two classes to have not


٤۱

dissimilar sizes; others assume that these classes follow a Gaussian distribution, etc [24].

Thresholding is one of the simplest and most widely used image segmentation techniques. The goal of thresholding is to segment an image into regions of interest and to remove all other regions deemed inessential.

The simplest thresholding methods use a single threshold in order to isolate objects of interest. In many cases, however, no single threshold provides a good segmentation result over an entire image. In such cases variable and multilevel threshold techniques based on various statistical measures are used.

The global thresholding technique is used to isolate objects of interest having values different from the background. Each pixel is classified as either belonging to an object of interest or to the background. This is accomplished by assigning to a pixel the value 1 if the source image value is within a given threshold range and 0 otherwise [26].

2.5.3 Histogram Thresholding

Histogram thresholding is one of the widely used techniques for monochrome image segmentation [25].

1. Histogram Smoothing

If the histogram is clearly bimodal, it is easy to find an appropriate threshold value. If the histogram contains multiple minima we can apply an algorithm for smoothing the histogram until it only contains only one minimum, which then becomes our threshold value. The algorithm runs in several steps. First, apply a derivative filter to the histogram. Then, repeatedly smooth until you are left with one minimum. The algorithm is outlined in Figure (2.5).


٤۲

2. Midpoint Method The midpoint method finds an appropriate threshold value in an

iterative fashion. First, apply a reasonable initial threshold value. Then, compute the mean of the pixel values below and above this threshold, respectively. Finally, compute the mean of the two means and use this value as the new threshold value. Continue until the difference between the two successive threshold values is smaller than a preset minimum. The algorithm is outlined in Figure (2.6).

Figure (2.5) Histogram smoothing Figure (2.6) Mid-point threshold method


٤۳

3. Minimum Error Method The minimum error method is similar to the midpoint method but

mostly gives a more precise result. First, apply a reasonable initial threshold value. Then, fit a Gaussian to the pixel values below and above this threshold, respectively. Then proceed as for the midpoint method. The algorithm is outlined in Figure (2.7). 4. The High-Pass Masking Method

First, apply a Laplacian filter to the input image f(x). Then, apply a threshold value to the output image g(x). A new histogram, Hist (g) is computed from the pixels, whose absolute values are higher than the threshold value (t), i.e. the thresholded g(x) is used as a mask applied to f(x). The masked pixels hopefully create a bimodal histogram from which the final threshold value (T) can be computed. The algorithm is outlined in Figure (2.8).

Figure (2.8) The high–pass masking method Figure (2.7) Minimum error method


٤٤

2.6 Brightness-Contrast Enhancement Image enhancement involves taking an image and improving it visually, typically by taking the advantage of the human visual system's response [18].

The principle of enhancement techniques is to process a given image, so that the result is more suitable than the original image for specific application [19].

Image enhancement techniques are used to emphasize and sharpen image features for display and analysis. Image enhancement is the process of applying these techniques to facilitate the development of a solution to a computer imaging problem. The range of applications includes using enhancement techniques as preprocessing steps to ease the next processing step or as postprocessing steps to improve the visual perception of a processed image, or image enhancement may be an end in itself.

The type of techniques includes point operations where each pixel is modified according to a particular equation that is not dependent on the other pixel values; mask operations, where each pixel is modified according to the values of the pixel's neighbors (using convolution mask); or global operations, where all pixel values in the image (or sub image) are taken into consideration [18].

2.6.1 Histogram Stretching One of the simplest and often most dramatic techniques is to simply stretch the contrast of an image. Histogram stretching is a form of gray-scale modification, some times referred to as histogram scaling.

The mapping function for a histogram stretch can be found by the following equation:

[ ] MINMINMAXcrIcrI

crIcrIcrIstretch +-´úû

ùêë

é-

-=

minmax

min

),(),(),(),()),(( )4.2.........(..........


٤٥

Where

max),( crI is the largest gray-level in the image I(r, c)

min),( crI is the smallest gray-level value in I(r, c) MAX and MIN correspond to the maximum and minimum gray-level

values (for an 8-bit image these are 0 and 255)

This equation will take an image and stretch the histogram across the entire gray-level range, which has the effect of increasing the contrast of a low contrast image. If a stretch is desired over a smaller range, different MAX and MIN values can be specified.

If most of the pixel values in an image fall within a small range, but a few outliers force the histogram to span the entire range, a pure histogram stretch will not improve the image. In this case it is useful to allow a small percentage of the pixel values to be clipped at the low and high end of the range, for an 8-bit image this means truncation at 0 and 255 [18].

The intensity mean and variance (or standard deviation) are two such properties frequently used in image processing. That is, the mean is a measure of average brightness and the variance is a measure of contrast [19].The mean and variance can used to determine MAX and MIN correspond to the maximum and minimum gray-level values where:

variance51meanmaxvariance51meanmin

´+=´-=

..

).......(.......... 52


٤٦

2.7 Chain Coding Chain coding is one of the popular error free coding techniques for binary images. This technique has been developed by Freeman not to be produce as an efficient code in sense of minimizing the number of codes bits required to describe a boundary, but rather to make a certain manipulation operations convenient.

Since digital images are usually acquired and processed in a grid format which follow equal spacing in the x and y directions, one could generate a chain code by following a boundary in, say a clockwise direction and assigning a direction to the segments connecting every pair of pixels.

The chain codes are used to represent a boundary by a connected sequence of straight line segments of specified length and direction. Typically this representation is based on the 4- or 8- connectivity of segments, where the direction of each segment is coded using a numbering scheme such as the one shown in Figure (2.9).

The chain can be regarded as concatenation of many small vectors where each vector is limited to one of the basic eight directions.

A useful feature of the chain notation is that a chain possesses a direction. This make it possible not only to describe the shape of the boundary but also to indicate in which direction the boundary was generated for those applications where this is interest [19].

y-1 y

y+1

x-1 x x+1

0

1

2

3

7 6

5

4

x-1 x x+1

y-1 y

y+1 0

1

3

2

a: 4-directional chain code. b: 8-directional chain code. Figure (2.9) Chain Coding.


٤۷

2.8 Pattern Recognition Pattern recognition has come to mean the study of artificial, as well

as, natural mechanisms that analyze, detect, recognize, and describe patterns in sensory and\or numerical data. Pattern recognition in fact, has grown into an interdisciplinary field that includes areas of engineering, computer science, biology, psychology and medicine [28].

The ease with which we recognize a face, understand spoken words, read hand written characters, identify our keys in our pocket by feel and decide whether an apple is ripe by its smell belies the astounding complex processes that underlie these acts of pattern recognition.

Pattern Recognition is the act of taking in raw data and taking an action based on the ‘category’ of the pattern. Pattern recognition has been crucial for our survival, and over the years a lot of researches were conducted to develop algorithms that duplicate the amazing ability of humans to recognize patterns [29].

2.8.1 The Pattern Recognition Components

1. Problem data: it is concerned with the representation of the input data, which can be measured from the objects to be recognized. A Feature vector is one method to represent an image, or a part of image (an object), by finding measurements on asset of features. The feature vector is an n-dimensional vector that contains these measurements [18, 29].

2. Feature extraction module: has the purpose of extracting (or collecting) some important information for the task at hand. In image processing the goal of image analysis is to extract information useful for solving application based problem. This done by intelligently reducing the amount of image data. Feature extraction is apart of data reduction process and followed by feature analysis. Exactly what we do with the features will be application dependent [18, 29].


٤۸

3. Feature selection module: has the purpose of extracting the features that are important to achieve the objective of interest. One of the important aspects of feature analysis is to determine exactly which features are important. After selection the required features, the feature vector is used to represent the image by finding measurement on asset of features. The feature vector is n-dimensional vector contains these measurements. When selecting features for use in computer imaging application, the most important factor to be concerned is the robustness of the features. A feature is robust if it will provide consistent result across the entire application domain.

4. Classifier module: has the purpose of classifying the data relying on the information conveyed by the selected features [29]. Figure (2.10) shows simple pattern recognition components.

Problem Data Feature Extraction Feature selection

Pattern Classification Pattern Classes

Figure (2.10) Components of a Pattern Recognition System.

Pattern raw data


٤۹

2.8.2 Distance Measurements There are different types of distance measurement that used to

recognize a pattern or classify the pattern class, the distance measurement are used as a measure of similarity.

Let ),,,(),,,(

21

21

nn

nn

yyyyxxxx

K

K

(a) Euclidean Distance: 21

)( in

i i yxD -= å = )6.2........(..........,.........

(b) City Block Distance: å=

-=n

iii yxD

1

)7.2........(..........,.........

(c) Hamming Distance: { }iiD yxn1i

-=

=L

max ).........(..........,......... 82

The definition of distance is quite arbitrary, similarity is also ill-defined quantity; it is difficult to think of any measure of similarity than subjective assessment. For the moment, we shall assume that any two patterns within the same class are "similar" and that they are "dissimilar" if they belong to different classes. Then our rule states that we are likely to find that the distance between vectors corresponding to patterns from the same class are small. That is they are likely to be small compared to the distances between patterns from different class [28].

2.9 Pattern Recognition in Image Processing

Different kinds of pattern recognition systems are based on visual data. Most of these pattern recognition applications use the methodology of image processing to handle the visual data and extract the required discriminating features for recognition.


٥۰

2.9.1 Scene analysis recognition Scene analysis starts with the extraction of important features from a

given visual data. The data could represent the physical world. That may be presented by image of many gray levels. The process to convert the physical object to data is by taking an image by imaging devices. These devices could be sensitive to visible light, x-ray, ultra violet, infra red, or a high frequencies system like radars. Once the image is taken by the image device then the image must converted to a digital form to enable the computer systems make some preprocessing steps. Sometimes enhancement or compression is necessary. All the previous steps are preprocessing steps. After that the image segmentation is an important part. The image segmentation may depend on the geometrical shapes, gray levels or the distance between the regions. The threshold is important to segment the image.

Once the image is segmented then the main features can be extracted like the moment relation between the segments, Fourier transform, Laplace transform or the correlation between them etc. These features can be considered as a property for the used object. The information can be obtained from the extracted features. These features can be used to classify the objects. Figure (2.11) shows the scene processing steps. Table (2.2) shows some examples of tasks to recognize the outputs from the inputs.

Task of each class Input data Output data Character recognition Optical signal Name of character

Speech recognition Acoustic wave forms Name of word Speaker recognition Voice Name of speaker Weather prediction Weather maps Weather forecast Medical diagnosis Symptoms Disease

Stock market predication

Financial news and charts

Prediction of market

Table (2.2) Example of Pattern recognition tasks with their types of input and the output data.


٥۱

Figure (2.11) The recognition steps.

Image Classification

Recognition

Decision

Pre-processing of Image

Imaging device

Image Segmentation

Feature Selection

Image Registration

Image Creation

Image Enhancement

Image Restoration

Physical Object

Digital Computer


٥۲

2.9.2 2D Recognition by Moment

One of the most important methods of object recognition is the moment method which is introduced by Hu. This method is used to identify 2D objects from 2D image, images of optical, x-ray, infra red, ultra violet or ultra sonic cameras, etc.

The main problems of 2D object recognition are the size changing, translation of position, rotation by an angle from the principle axes and reflection about the x axis or about the Y axis or vice versa.

In addition to the above problems there are some other problems like colors or gray levels and the shadow effects. The size changing problem can be solved by many ways like normalizing the area of the object to 1; this normalization must be applied to the prototypes and the models.

The translation can be normalized by shifting the origin coordinates (0, 0) to the center of gravity of the object. After that the translations has no effect any more.

The rotation effect can be solved by detecting the orientation angle of the object with the horizontal axis or the vertical axis. There are many techniques of using the moments approaches; statistical moments, affine moments, and the orthogonal moments.

Recently a test of radial and angular invariant moments have been specified which also are invariant to all the variations of size, translation, rotation and reflections.

2.9.2.1 Pattern Recognition by Moment Invariants

The pattern recognition of an object is very important features to analysis images of a given scene containing many objects (each region can be treated alone). The main important thing is how to get the necessary features from an image by analyzing it.


٥۳

One of important points is how to use the pattern recognition techniques to analyse the images. A good example of that is how computer recognize its vision or (a vision scene).

Actually the recognition is done by analyzing the vision in four steps. The first step is converting the scene into array of numbers representing each digital pixel by specific numbers representing location and intensity of colors (or gray level).

The second step is to enhance the image by removing the noise. And if necessary make segmentation which means separating each region in sets. The separation can be done by segmentation like the size of similar intensity of regions, geometrical size of each region or by the location of each region (clustering) and minimum distance.

The third step is the features extraction which means how to get the necessary information from the scene which represent the elimination of the least important features to analyse any region or any image of an object.

In the fourth step the sizes, locations and the orientations must be normalized, this technique is used to recognize the deformed object by size changing, translation, reflection and rotation. This approach is based on the theory of algebraic invariants. The invariants of the second and the third order moments are derived by Hu. During the transformation the features of the images may be changed, the changes may be normalized.

2.9.2.2 Moment In General

In general, moments describe numeric quantities at some distance from a reference point or axis. The operation of the image feature vector extraction by moments is one of the common techniques used these days, where each moment order has different information for the same image.

These moments are also divided into orthogonal (e.g. Zernike and Legendre), non orthogonal such as Regular moments, and Complex moments.


٥٤

The theory of moments provides an interesting and sometimes useful alternative for series expansions for representing objects and shapes.

The use of moments for image analysis is straightforward if we consider a binary or gray level image segment as a two-dimensional density distribution function. In this way, moments may be used to characterize an image segment and extract properties that have analogies in statistics.

Operations such as rotation, translation, scaling, and reflection may exist for each image. They are also called transformations. These transformations cause changes for each order of image moments. The solutions were introduced to keep the moments constant or invariants, which are called moment invariants [30].

2.9.2.3 Moment Invariants

Moment invariants have become a classical tool for object recognition during the last 30 years. They were firstly introduced to the pattern recognition community by Hu in 1962; who employed the results of the theory of algebraic invariants and derived his seven famous invariants to the rotation of 2-D objects [19].

The theory of algebra invariants are used to drive affine moment invariants:

yaxaau 210 ++= , )9.2.........(..........

ybxbbv 210 ++= , )10.2........(..........

Given a two dimensional continuous function f(x, y) we define the moment of order (p+ q) by the relation

ò ò¥

¥-

¥

¥-

= dxdyyxfyxm qppq ),( ,....2,1,0, =qp , )11.2........(..........

where p, q are the order of the invariant moments about the x axis and the y axis respectively.


٥٥

A uniqueness theorem states that if f(x, y) is piecewise continuous and has nonzero values only in a finite part of the x-y plan, then moments of all orders exist and the moment sequence ( pqm ) is uniquely determined by f(x, y).

If we use central moments instead of the general moments (2.11) any function of them will be invariant under translation. Conversely ( pqm )

uniquely determines f(x, y).The central moments pqm are defined as:

dxdyyxfyyxx qppq ),()()( --= ò ò

¥

¥-

¥

¥-

m , )12.2........(..........

,....2,1,0, =qp

00

10

mmx =

00

01

mm

y =

Where yx, are the coordinates of the center of gravity of a given object.

Figure (2.12) illustrates the measurement center of gravity of mass distribution.

Center of gravity ),( yxG

X

Y

y

x )0,0(

Figure (2.12) The normalization of translation.


٥٦

For a digital image, Eq. (2.12) becomes

),()()( yxfyyxx q

x y

ppq --= ååm )13.2....(..............................

The central moments up to order 3 are as follows:

0),()( 1

10 =-= åå yxfxxx y

m )14.2(....................,......... a

0),()(01 =-= åå yxfyyx y

m )14.2(....................,......... b

),()()(11 yxfyyxxx y

--= ååm00

011011 m

mmm -= )14.2(....................,......... c

),()( 2

20 yxfxxx y

åå -=m 00

210

20 mm

m -= )14.2(....................,......... d

),()( 202 yxfyy

x yåå -=m

00

201

02 mm

m -= )14.2(....................,......... e

),()( 3

30 yxfxxx y

åå -=m 2102030 23 xmmxm +-= )14.2(....................,......... f

),()()( 2

12 yxfyyxxx y

--= ååm 102

021112 22 mymxmym +--= )14.2.....(,......... g

),()()( 2

21 yxfyyxxx y

--= ååm 012

201121 22 mxmymxm +--= )14.2....(,......... h

åå -=

x yyxfyy ),()( 3

03m 012

0203 23 mymym +-= )14.2.....(..........,......... i

In summary,

0000 m=m , 101111 mym -=m

010 =m , 210203030 23 xmmxm +-=m

001 =m , 102

02111212 22 mymxmym +--=m

102020 mxm -=m , 012

20112121 22 mxmymxm +--=m

010202 mym -=m , 012

020303 23 mymym +-=m


٥۷

The normalized central moments, denoted by pqh , are defined as

mm

h g00

pqpq = )15.2......(..........,.........

where 12

++

=qpg )16.2......(..........,.........

For ,......3,2=+ qp

From the second and third moments, a set of seven invariant moments can be derived. They are given by

hhf 02201 += )17.2.(..............................,......... a

( ) hhhf 211

202202 4+-= )17.2..(..............................,......... b

( ) ( )20321

212303 33 hhhhf -+-= )17.2.(..............................,......... c

( ) ( )20321

212304 hhhhf -+-= )17.2..(..............................,......... d

( )( )( ) ( )[ ]( )( ) ( ) ( )[ ]2

03212

123003210321

20321

21230123012305

33

33hhhhhhhh

hhhhhhhhf+-++-+

+-++-= )17.2..(..........,......... e

( ) ( ) ( )[ ]20321

2123002206 hhhhhhf +-+-= )17.2..(..............................,......... f

( )( ) ( ) ( )[ ]( )( ) ( ) ( )[ ]2

03212

123003213012

20321

21230123030217

33

33

hhhhhhhh

hhhhhhhhf

+-++-+

+-++-= )17.2...(..........,......... g

This set of moments has been shown to be invariant to translation, rotation, and scale change [19].

The above seven moments vary considerably; the last two might varied larger than other moments. To grant each moment invariant an equal weight, they need to be normalized such that they cover approximately the same range [27].


٥۸

The moments ),(71 ff are called Moment invariants. They are rotation,

scale and translation invariant and very useful to represent shapes and objects in pattern recognition problems [32].

It is important to refer to the following remarks related to moments types:

1. Central moments are not invariant to rotation.

2. Regular moments change if object is not at the same location.

3. Central moments are invariant to translation.

2.10 Complex Moment

As another type of moments is the complex moments. This set of moments was conducted to:

1. Simplify the moments determination.

2. Get moments set which are more invariant to translation, rotation, and scaling.

The simplest set of complex moments could be defined as follows:

åå ¢+¢=y x

m yxfyjxmM ),()()( )18.2..(..........,.........

where 1-=j

yx ¢¢, are the normalized coordinates, which can be determined by using on of the following equations:

1st Set:

c

c

xxx

x-

=¢ )19.2.........(,......... a

c

c

yyy

y-

=¢ )19.2.........(,......... b


٥۹

where

2ImgWid=cx )20.2.........(,......... a

2ImgHgtyc = )20.2.........(,......... b

2nd Set:

x

xxx -=¢ )21.2.........(,......... a

y

yyy -=¢ )21.2.........(,......... b

where

åå åå=y x

yxfyxxfx ),(),( )22.2.........(,......... a

åå åå=y x

yxfyxyfy ),(),( )22.2.........(,......... b

Since the complex moments are complex, so they consist of the real )( rM and imaginary )( iM parts. The first four complex moments could be

expressed as follows:

åå=x y

r yxxfM ),(1 , åå=x y

i yxyfM ),(1 )23.2.........(,.........

( ) ),(222 yxfyxMx y

r åå ¢-¢= , åå ¢¢=x y

i yxfyxM ),(2 )24.2.........(,.........

( ) ),(3 233 yxfyxxM

x yr åå ¢¢-¢= , ( ) ),(3 323 yxfyyxM

x yi åå ¢-¢¢= )25.2.........(,.........

( ) ),(6 42244 yxfyyxxM

x yr åå ¢+¢¢-¢= , ),(234 yxfyxyxM

x yi åå ¢¢-¢¢= )26.2......(,.........

Chapter three Palmprint Authentication System

٦۰

Chapter Three Hand Palmprint Authentication System

3.1 System Description This chapter is dedicated to describe the designed and implemented

hand palmprint authentication system. The designed system consists of the following steps:

1. Image acquisition: Obtain hand image by using any imaging device,

in the implemented system images were taken using scanner, The snap shot images must be in Bmp format with resolution 100 dpi, which is so suitable to extract principal lines, wrinkles line. During the image acquisition process the following steps should be fulfilled:

a. Place the hand on the flat surface of scanner device, do not press the hand so tightly, and make sure your palm is just touching the scanner surface.

b. Keep your fingers separated by reasonable space, and make sure that your fingers not so far from one to another.

c. The wrist must not placed; put the arm should be perpendicular to the hand.

2. Preprocessing: Perform all required digital image processing steps to specify the palmprint area from the hand image.

3. Feature extraction: find some stable and unique discriminating features. Feature extraction is a key issue for pattern recognition, two types of moments used to represent the central palmprint image, the Moment Invariant and Complex Moment.

4. Enrollment Phase: save the extracted features from the training establish the reference features vector for each registered person.


٦۱

5. Verification Phase: match the extracted features with the corresponding features extracted from the same person that is previously enrolled in the system.

Figure (3.1) Original Image


٦۲

3.2 Image Acquisition The goal of this step is to get the width and the height of the input

image and to decompose the RGB colors in a separate 3 color components array. The image file consists of header and image data. To read the Bmp image format the process should start with reading the header information from the bitmap image file. The header contains information such as width, height, the file type, the number of bit per pixel, and other information that mentioned in Table (2.1).

The data structure that is used to represent the header information is selected to be a record named as Bmpheader, the structure of this record is shown below:

Signature: of type char [2] "BM" File size: of type long R1, R2: of type integer Offset Position of type Long Size of bitmap_infoheader of type Long Width of type Long Height of type Long nplane of type Integer nbp of type Integer Compression of type Long sizepad of type Long HR of type Long VR of type Long ncolor1 of type Long ncolor2 of type Long

Where

· Width: image width. · Height: image height. · nbp: number of bits per pixel.


٦۳

Algorithm (3.1) Load BMP Image Input: Bmp image file of 24-bits/pixel. Output: Header, RGB colors each color component is in separate array. Begin

1. Open the Bmp image for binary reading 2. Get the header information and store the contents in the

Bmp header record. 3. Get the width (Wid), height (Hgt), nbp from the Bmp header

record. 4. If nbp=24 then 5. W=trunc (Wid*24+31)/32) * 4 6. y=Hgt -1 7. Get raw by raw from the file, and do 8. x=0, xx=3* x 9. blue(x,y)=a(xx), green(x,y)=a(xx+1), red(x,y)=a(xx+2) 10. x=x+1 11. if x<= Wid then xx=3*x and goto 9 12. y=y-1 13. if y>=0 goto 7 14. Close the file.

End Where

· blue, red, green array of two dimension (0 to Wid-1, 0 to Hgt-1) and of byte data type.

· The (a) array of one dimension of byte data type, which stores the raw from the file.


٦٤

3.3 Palmprint Image Extraction Preprocessing The goal of preprocessing is to extract the palmprint image area.

3.3.1 Gray Image Convert an original color image to gray-scale image; there are a

variety ways to produce the gray-scale image using the RGB color image.

1.3

yxByxGyxRyxGray ),(),(),(),( ++=

2. )),(),,(),,(max(),( yxByxGyxRyxGray =

3. ),(),( yxRyxGray = , ),(),( yxGyxGray = , ),(),( yxByxGray =

3.3.2 Binary Image

We will convert the gray image to binary image using thresholding technique. By testing the three above mentioned ways of conversion gray image to binary image for palm segmentation process, it is found that the third method ),(),( yxRyxGray = is more suitable one, while other methods cause the disappearance of some parts of the palm image after the binarization.

To convert the gray-scale image to binary image we need to get the suitable threshold value .This will require of calculation histogram, since the histogram of the hand image is bimodal, so it is useful to determine the threshold value and makes the threshold determination operation easier. This method will let the determined threshold value depends on the histogram of each captured image.


٦٥

The Major Steps of Binarization Process

1. Calculate the image histogram: For each gray-level value computes the number of pixels has that gray value. Let His (0-255) is a vector consists of 256 entry, each entry register the frequency of occurrence of a distinct gray level. Let the image height (Hgt) and the image width (Wid).

Algorithm (3. 2) Histogram Determination Input: Gray(x, y) array, Width, Height of the gray image Output: image histogram His (0 to 255) Begin 1. Set k=0, x=0, y=0 2. His (k) =0, increment k counter by one 3. If k<256 goto 2 4. Scan the gray image Gray(x, y) raw by raw 4.1 Set k=Gray(x, y), His (k) =His (k) +1

4.2 x=x+1: if x<wid then Goto 4.1 4.3 x=0, y=y+1 4.4 if y<=hgt-1 goto 4 End.

2. Finding the threshold: The bimodal histogram is a histogram with two major peaks. Find the position of the two major peaks p1, p2. (i.e. the gray level values that have most repeatedly occurrence in the image.

Algorithm (3. 3) Threshold Determination

Input: image histogram His (0 to 255) Output: T threshold value Initial: i=0, max1=0, p1=0, max2=0, p2=0 Begin Get the 1'st major peak

1.1 if His (i)>max1 then max1=His(i) , p1=i

1.2 i=i+1, if i<=127 then goto 1.1


٦٦

Get the 2'st major peak 1.3 i=128 1.4 if his(i) >= max2 then max2=His(i) , p2=i 1.5 i=i+1,if i<=255 goto 1.4 Get the threshold

1.6 2

21 ppT +=

End. 3. Using the threshold value to convert the gray-image to binary image.

Let Gray(x, y) is the gray image, T is the threshold value, and Bin(x, y) is the binary image.

Algorithm (3.4) Binarization by threshold Input Gray(x,y) ,Width, Height of the input image, threshold T. Output: Binary image Bin(x, y) Begin 1. Initialize: y=0 2. For each Gray(x, y) in the image 2.1 If Gray(x, y) > T then Bin(x, y) = 1 else Bin(x, y) = 0 2.2 x=x+1: if x<Wid then goto 2.1 2.3 x=0, y=y+1 2.4 if y<=Hgt-1 then goto 2.1

End.

Figure (3.2) Binary image


٦۷

3.3.3 Smooth the Binary Image Apply a low pass filter such as Mean filter on the binary image, to

eliminate the noise from the binary image that caused during the capturing of the image or due to binarization by the thresholding. Also this process will smooth the boundary of the binary image.

By using convolution process that require us to overly the mask on the image multiply the corresponding mask values by pixels values, and sum all these multiplication results. The process continues until we reach the end of the raw, each time we place the result of the filter operation in the location corresponding to the center of the mask. When the end of the raw is reached, the mask is moved down one raw, and the process is repeated raw by raw until this procedure is performed on the entire image, the process of sliding, multiplying and summing is called convolution. Note that the output image must be put in separate image array.

Algorithm (3.5) Smooth the Binary Image by Mean Filter

Input: Binary image Bin(x, y) Output: Smoothed Binary image Binm(x, y)

Begin 1. set y=1 2. x=1 3. Sum=0, r=-1 4. c=-1 5. Sum=Sum+ Bin(r+c, c+y) 6. c=c+1 ,if c<=1 then goto 5 7. r=r+1 if r<= 1 the goto 4 8. if sum>2 then Binm(x, y) =1 else Binm(x, y) =0 9. x=x+1, if x<=wid-2 then goto 3 10. y=y+1,if y<=hgt-2 then goto 2

End


٦۸

The pixel's values of the Binary image after smoothing must be converted to be either 255 or 0. This can be done by applying the following condition on all pixels of the image:

If Binm(x, y) =1 then Binm(x, y) =255 for all values of x and y.

3.3.4 Image Boundary

The boundary (contour line of one pixel thickness) of the binary image could be allocated by using algorithm (3.6).

Algorithm (3.6) Image Boundary tracing Input: Binary image Binm(x, y), white are to the hand area and

black in the back ground area. Output: Image Boundary P, i.e., a 2dimensional array consist two

values: 0 corresponds to back ground and 255 corresponds to boundary and nx ( ), ny ( ) are two one dimensional arrays that store the x, y coordinates, respectively, of the boundary pixel value.

Figure (3.3) Smooth binary image


٦۹

1. Scan the Binary image raw by raw (from the left to right, or from right to left.) to find the transition from black to white. (i.e., Binm(x, y) = 0 and Binm(x+1, y) =255).Or transition from white to black. (i.e., Binm(x, y) =255 and Binm(x+1, y) =0).

Then store the x-coordinate and y-coordinate of the white pixel in the array x1( ), y1 ( ).

2. Scan the Binary image column by column (from the left to right, or

from right to left.) to find the transition from black to white (i.e., Binm (x, y) = 0 and Binm (x, y+1) =255).Or transition from white to black (i.e., Binm(x, y) =255 and Binm(x, y+1) =0).

Then store the x-coordinate and y-coordinate of the white pixel in the array x2( ), y2 ( ).

The output of this stage are two pairs of arrays, each array contain the coordinates of some boundary pixels. After sorting and merge the two arrays in Nx ( ), Ny ( ), removing the existing repetition in some pixels value then we will get the image boundary array (Nx ( ), Ny ( )).

Then the boundary arrays should sorted in ascending order according to x or y values. After this step we need to represent a boundary by a connected sequence of straight line segments of specified length and direction. This is done by using the chain coding algorithm (which will be discussed in the next step).

Figure (3.4) Image boundary


۷۰

3.3.5 Chain Coding In previous step we get the image boundary sorted according to x

or y-coordinate values, but we need to represent the boundary as a connected sequence, this is done by using the Chain Coding algorithm.

Algorithm (3.7) Chain Coding

Input: Image Boundary P, array consist two value 0 corresponds to back ground and 255 corresponds to boundary, Nx ( ), Ny ( ), Wid, Hgt.

Output: Chain code, Chx (0 to r-1), Chy (0 to r-1) and r is the number of boundary pixels.

Begin Initialization

1. i=0, x=Nx (0), y=Ny (0).

2. if pixel value P(x, y) =255 then store the coordinate Chx (i) =x,

Chy (i) =y, i=i+1, and set the pixel value to black(P(x,y)=0)to prevent returning back to it.

3. To prevent search out of range(out of image dimension), the(x) and (y) values that less than zero, set to zero, the (y) values that more than image height then set to Height-2,and the (x) values that more than image Width then set to Width-2.

That is:

Set Xp = x + 1: If Xp >= wid Then x = wid - 2 Set Xm = x - 1: If Xm < 0 Then x = 1 Set Yp = y + 1: If Yp > = hgt Then y = hgt - 2 Set Ym = y - 1: If Ym < 0 Then y = 1


۷۱

4. Search around the P(x, y) neighbors, the nearest white pixel in a

clock wise direction shown in figure (2.9.a). Then for that pixel find the nearest pixel and so on until no more white pixel will detected.

4.1 If p (Xp, y) = 255 Then x = Xp, goto 1.

4.2 If p (Xp, Yp) = 255 Then x = Xp: y = Yp, goto 1.

4.3 If p (x, Yp) = 255 Then y = Yp, goto 1.

4.4 If p (Xm, Yp) = 255 Then x = Xm: y = Yp, goto 1.

4.5 If p (Xm, y) = 255 Then x = Xm, goto 1.

4.6 If p (Xm, Ym) = 255 Then x = Xm: y = Ym, goto 1.

4.7 If p (x, Ym) = 255 Then y = Ym, goto 1.

4.8 If p (Xp, Ym) = 255 Then x = Xp: y = Ym, goto 1.

Store the number of chain code pixels in r

5. r = i.

End.

Figure (3.5) Chain Boundary


۷۲

3.3.6 Palmprint Area detection System

It is important to isolate the palmprint from other parts of the hand image. This step is very important to improve the palmprint recognition process. The three major steps in determining the coordinate systems are:

3.3.6.1 Detection of Finger's Join points and End points

The join points are the nodes joining the fingers of each hand, there are four join points, and three end points.

The join point algorithm uses the chain codes, which are boundary as a connected sequence of boundary points. This will make the join points detection operation easy.

Using the fact that each join point has lower Y value relative to its surrounding neighbors (i.e. neighbors from both sides left and right). But join point region sometimes contain more than one point that has local lower Y value, i.e. in the case that the join region is a flat area.

Thus to get it we must search around the point in the chain to check whether it has the lower y value relative to its neighbors.

3.3.6.1.1 Detection of Finger's Join points

Two passes to are required to find the join points

I.First pass: Find the join point regions, there are four regions, one region to each join point, each region contains a number of points that satisfy the above condition, scan the chain (point by point), such that at each point compute the mean value of 20 points from preceding the tested point in the chain, and also determine the mean of the Y coordinates of the points coming after the tested point. If the point Y coordinate value is lower than the two mean y-values of the previous and next points, then save the tested point value as a join point region.


۷۳

Algorithm (3.8) Detection the Joinpoints Region

Input: Chain boundary array, Chx ( ), Chy ( ), r is a number of pixels in the chain

Output: Prx ( ) the x-coordinates of the join region, Pry ( ) the y-coordinates of the join region, t is the number of pixels in the join regions.

Begin 1. set i=0, t=0 2. Scan the chain boundary pixel by pixel, read from chain code

Chx(i),Chy(i) 3. set Sl=0, Sr=0 4. Compute the mean value for the 20 pixels in the both sides before

and after the current tested pixel (Chx (i), Chy (i)). The mean for the y-coordinates values could be determined as follows:

4.1 for j=1 to 20 4.2 Sl = Sl + Chy (abs (i-j)), Sr =Sr + Chy (i+j) 4.3 end for 4.4 Ml=round (Ml / 20), Mr=round (Mr / 20) If (Chy (i) < = Mr and Chy (i) < = Ml) Then Prx (t) =Chx (i), Pry (t) =Chy (i), t=t+1 4.5 i=i+1 4.6 If i < = r - 20 goto 2

End

Figure (3.6) Join point region


۷٤

II. Second pass: From the join point region, select the middle point that represent the join point. This is done by computing the mean value for the 5 points from right, and 5 points from left side to Y value. If the Y-coordinate of the tested point is lower than the Y mean right and Y mean left, then save that point as join point. To get exactly the only four points (A, B, C, D) we must eliminate the points that has the Y values less than or equal 20 or the difference between the X-coordinate of the points is less than 11, all that to make sure that we will get join points only.

Algorithm (3.9) Detection the Joinpoints

Input: The arrays Prx ( ), Pry ( ), t is the number of pixels in the join regions.

Output: Pjx ( ) the x-coordinates of the join points, Pjy( ) the y-coordinate of the join points, s is the number of pixels in the regions.

1. i= 0, s = 0 2. Read PRx(i) , Pry (i) 3. set Sl=0 , Sr=0 4. Compute the mean value of the y-coordinates of the pixels in

the both sides (right and left) from the current tested pixel (PRx (i), PRy (i)). The mean of y-coordinate could be determined as follows: 4.1 For j = 1 to 5

4.2 Sl=Sl + Prx( abs( i- j) ), Sr=Sr + Pry( i+j ) 4.3 next j

5. Ml = round ( Ml/20 ) , Mr=round (Mr/20) If (Pry (i)) < Mr and Pry (i) < Ml)

Then Pjx (s) =Prx (i), Pjy (s) =Pry (i), s=s+1 6. i=i+1 7. If i < = t - 5 goto 2


۷٥

8. Then eliminate the pixels whose x-coordinate is very close to each other, and keep only one of them value. Also eliminate the pixels that has the y-coordinates less than or equal 20. The previous steps of join points detection may find some points near to wrist which must be ignored.

8.1 set i=0, j=0

8.2 Read Pjx(i), Pjy(i), set j=i+1

8.3 If abs ( Pjx (j) - Pjx (i) ) <= 10 or Pjy (i) <= 20

Then Pjx (i) =-1, Pjx (i) =-1

8.4 j=j+1 , If j <=s then goto 8.3

8.5 i=i+1 , If i<= s then goto 8.2

8.6 end

9. Sort the arrays Pjx( ),Pjy( ) according to x-values registered in Pjx( ).

A

B C D

Figure (3.7) Join points (A, B, C, D)

A

B C D


۷٦

3.3.6.1.2 Detection of Finger's End points The finger's end points can be defined as the point where the finger

is end. To allocate the end points the following steps were followed:

1. First end point (E): find it by searching the nearest point to D, where it x-coordinates distance should be more than 10,this nearest point should satisfy the line equation passing through C and D, and the x-coordinate of E higher than the x-coordinate of point C.

Algorithm (3.10) First End- point Estimation

Input: join point D as (x1, y1); join point C as (x2, y2); Chain code as Chx(0 to r), Chy(0 to r), r number of pixels in the Chain code.

Output: The end point E coordinates as (x3, y3), stored in (xy (0), xy (1)).

Begin

1. set i=0, min=9999, )12()12( xxyyss --=

2. Read the pixel in the chain, x=chx (i), y=chy (i).

3. If (x>x1) then

3.1. 22 )1()1( yyxxdis -+-=

3.2. dif=abs ( ss * (x-x1) + y1- y)

3.3. If (dif < min) and (dis >10) then x3=x, y3=y, min=dif

4. i=i+1, If I <= r- 1 then goto 2

5. Store (x3, y3) in the array xy(0) =x3, xy(1) = y3

End.


۷۷

2. Second end point (F): find it by searching the nearest point to B where the horizontal distance more than 15, such that it satisfies the line equation passing through C to B, and its x is less than the x-coordinates of both point C and B.

Algorithm (3.11) Second End-Point Estimation

Input: join point C as (x1, y1); join point B as (x2, y2); Chain code as Chx(0 to r), Chy (0 to r); r number of pixel in the chain code.

Output: The end point F as (x3, y3), stored in (xy (2), xy (3)).

Begin

1. set i=0, min=9999, )12()12( xxyyss --=


3. If (x<=x1) and (x< =x2) and (y<=y1) Then

3.1 22 )2()2( yyxxdis -+-=

3.2 dif=abs (ss*(x-x1) +y1-y)

3.3 If (dif < min) and (dis >15) and (dif >1) then

x3=x, y3=y, min=dif

4. i=i+1; If i <= r-1 then goto 2

5. Store x3, y3 in the array xy (2) =x3, xy (3) = y3

End.

3. Third end point (G): Its position could be estimated by searching for

the nearest point to A, such that its distance is more than ten, this end point should satisfy the line equation passing through A to F, and its x and y-coordinates should less than the x-coordinate &y-coordinates of the points A and F.


۷۸

Algorithm (3.12) Third End-Point Estimation

Input: join point A as (x1, y1); end point F as (x2, y2);Chain code as Chx(0 to r), Chy(0 to r); r number of pixel in the chain..

Output: The end point G as (x3, y3), store in (xy (4), xy (5)).

Begin

1. set i=0, min=9999, )12()12( xxyyss --=


3. If (x<x2) and (y< =y2) and (x<x1) and (y<y1) Then

3.1 22 )2()2( yyxxdis -+-=

3.2 dif=abs (ss*(x-x1) +y1-y)

3.3 If (dif< min) and (dis>10) then x3=x, y3=y, min=dif

4. i = i+1, If i<=r-1 then goto 2

5. Store x3, y3 in the array xy (4) =x3, xy (5) = y3

End.

After finding A, B, C, D, E, F and G points then uses them to extract the palmprint from the hand by remove the fingers from the image.

E

G

F

A

B D

C

Figure (3.8) Join points and End points


۷۹

3.3.6.2 Get the Palm Image Area To get palm image without the fingers, could be done by using:

(1) Image chain boundary.

(2) Join points and End points (A, B, C, D, E, F, G).

This segmentation could be done applying the following steps:

1. Scan the chain code to get the points starting from point G and reach the point E in direction shown in the figure (3.9)

2. Scan the chain code to get the points starting from point F and reach point A in the direction shown in the figure (3.9).

3. Save the all points collected in the above the two steps in a new array.

4. Evaluate a Mask that contains two values: 1's corresponds to palm points and 0's corresponds to back ground and fingers.

(a) Defined an array (palm mask) whose size is image size.

(b) Set the values of mask elements to zero.

(c) Set a picture pixel to black.

E

G

F

A

B D

C

Figure (3.9) The Direction from G to E and the direction from F to A


۸۰

5. Put in the picture the points of the array (pp) as white points connect the join points and end points by drawing lines between the points G&A, F&B, B&C, C&D, D&E shown in the figure (3.10). In this step we will get a closed contour that specifies the palm area only.

6. Compute the palm center; fill the inner region by a white color shown in the figure (3.11).

7. Scan the picture from (0, 0) to (wid-1, hgt-1), If Point(x, y) is white it is considered as palm point then set mask(x, y) =1.

8. Then get the red, green, blue components values of the palm points only (from the original image array). This could be done as follows: for all pixels in the image:

If mask(x, y) =1 then Palmr(x, y) = red(x, y) Palmb (x, y) =blue(x, y) Palmg (x, y) =green(x, y) else

Palmr(x, y) =0 Palmb(x, y) =0 Palmg(x, y) =0

Figure (3.11) The Palm Area

A

Figure (3.10) lines between the points G&A, F&B, B&C, C&D, D&E

F B C D

E

G

A


۸۱

3.3.6.3 Shifting the coordinate system

The palm image in the previous step has the same coordinate of the original image, using this coordinate in the next is manipulation not suitable. The palm area must shift to (0, 0), get new width (Wn) and new height (Hn) this done by applying the following steps:

1. Create a widow that surround the palm, i.e. finding MaxX, MinX,

MaxY, MinY to the palm region only shown in the Figure (3.12).

2. compute Wn and Hn : Wn= MaxX- MinX, Hn=MaxY - MinY

3. Shift the palm by (-MinX,-MinY), and create new Palm array :

3.1 for all points in the palm color array

3.2 If mask(x, y) =1 then npalmr(x-MinX, y-MinY) = Palmr(x, y)

npalmb (x-MinX, y-MinY) =Palmb(x, y)

npalmg (x-MinX, y-MinY) =Palmg(x, y)

Palm shift mask(x-MinX,-y-MinY) = 1

The output of this step: is the three color (R, G, and B) component contents of the palm area only, the arrays npalmr, npalmg, npalmb with dimension Wn and Hn, and the Palm shift mask will specify the palm points only.

Width

MaxX MixX

MinY

MaxX

Figure (3.12) a widow that surrounds the palm

Height

Wn

Hn

Figure (3.13) Palm area with dimension Wn and Hn


۸۲

3.3.7 Palmprint Image Enhancement Applying brightness-contrast enhancement on the palmprint image is very useful to improve the extraction of the principal lines may exist in the palmprint area. This kind of enhancement could be done by using histogram stretching method discussed in section (2.6.1).The enhancement process should be done to the palm pixels only.

Algorithm (3.13) Brightness-Contrast Enhancement

Input: Image A(x, y), Wn, Hn.

Output: Enhanced Palmprint Image Enh(x, y)

Begin

1. Compute the Mean value of the palm image.

1.1 set y=0 ; sum=0; s=0

1.2 set x=0

1.3 If Palm shift mask(x,y)=1 then sum= sum+ A(x,y), s=s+1

1.4 x=x+1; If (x < = Wn ) then goto 1.3

1.5 y=y+1; If (y <= Hn) then goto 1.2

1.6 mean=sum/s

2. Compute the variance (standard deviation)

2.1 set y=0, sum=0

2.2 set x=0

2.3 If Palm shift mask(x,y)=1 then 2)),(( meanyxasumsum -+=

2.4 x=x+1, If x<=Wn then goto 2.3

2.5 y=y+1, If y<=Hn then goto 2.2

2.6 ssumsd =


۸۳

3. Stretching process

3.1 min= mean-1.5 * sd , max = mean+1.5*sd

3.2 set y=0

3.3 x=0

3.4 If A (x,y) < min then Enh(x,y) = 0 goto 3.7

3.5 If A(x,y) > max then Enh(x,y) = 255 goto 3.7

3.6 Enh(x,y)=((a(x,y)-min)/(max-min))*255

3.7 x=x+1; if x <= Wn then goto 3.4

3.8 y=y+1; if y <= Hn then goto 3.3

End

Where A is any of the color arrays npalmr, npalmg, npalmb (we have made different test and find out the suitable one is npalmg).

3.3.8 Edge Detection Using Sobel Operator

As presented in section (2.4.1) the Sobel masks are used to improve the detection of the lines in the palm. The raw and column Sobel masks are convolved with the image, this process done only on the palm pixels whose all neighbor belong to the palm region.

Figure (3.14) Palmprint Enhancement


۸٤

Algorithm (3.14) Sobel Operator

Input: Enhancement Palmprint image Ehn (0 to Wn, 0 to Hn), Wn, Hn.

Output: Sob (0 to Wn, 0 to Hn)

Begin

1. Call Sobel masks: Hsobel (-1 to 1)

, Vsobel (-1 to 1)

2. y=1

3. x=1

4. sum1=0 ; sum2=0

5. If all the pixels within the 3*3 window are in the palm region.

Then

5.1 r = -1

5.2 c = -1

5.3 sum1=sum1+Hsobel (r,c) * Enh( r + x,c + y)

5.4 sum2=sum2+Vsobel (r,c) * Enh(r + x, c +y)

5.5 c=c+1, If c < = 1 then goto 5.3

5.6 r=r+1, If r < = 1 then goto 5.2

5.7 22 21),( sumsumyxSobel +=

5.8 To improve lines multiply sobel result by two

Sobel(x, y) =Sobel(x, y)*2

6. Else set Sobel(x,y)=0

7. x = x + 1; if x < = Wn then goto 4

8. y= y + 1; if y < = Hn then goto 3

End.

Figure (3.15) Sobel operator on the Palmprint image


۸٥

3.3.9 Get Central Palmprint In order to get the central part of the palm we need to determine the center point of the palm region and it estimated radius.

The center of mass of the palm point could be determined by using the following equations:

åå

åå=

x y

x yc yxSobel

yxxSobelx

),(

),(,

åå

åå=

x y

x yc yxSobel

yxySobely

),(

),(

The radius of the palmprint could be determined by using the following equations:

åå

åå -=

x y

x yc

x yxSobel

yxSobelxx

),(

),()( 2

s

åå

åå -=

x y

x yc

y yxSobel

yxSobelyy

),(

),()( 2

s

22yxRadius ss +=

It also needs to determine the distance of each point in the palm from the center. For each point in the palm, if the distance less than the radius then store that point as point in the central part of the palm.

Figure (3.16) Central Palmprint


۸٦

3.4 Feature Extraction Feature extraction is a key issue for pattern recognition, two types

of moments used to represent the central palmprint image, the Moment Invariant and Complex Moment.

3.4.1 Feature Extraction Using Moments Invariant

As presented in section (2.9.2.3) moments invariant have become a classical tool for objects recognition. In this research work, the set of seven moments invariant was used to recognize the central palmprint image, where each moment order reflects different information for the central palmprint image, this process was done only on the pixels belong to the central palm region.

To determine the set of seven moment invariants we must follow the following steps:

1. Determine the weight of each pixel in the palm.

The values of moments are strongly affected by the position of the pixel, whenever the pixel distance from the center increase this pixel will significantly affected the moment value. Therefore to reduce this effect a weighting function is applied on each pixel, the weight value depends on the pixel distance from the center of palm. The weighting function will reduce the effect of pixels that are far away from the center to balance the effect of the nearest pixel to the center, this weighting step is adopted because our main interest is concentrated on the features of the central palm region.

The adopted weight function is

),(),( yxde12yxWeight a+

= ,…….…….. (3.1)

where

d(x,y) is the distance of the pixel from the palm center.

}.,....,.,.{ 0001000400050=a


۸۷

Algorithm (3.15) Determination of Pixel Weight

Input: Hn, Wn, Xc, Yc,a

Output: Weight(x,y) array.

Begin

1. set y = 0

2. x = 0

3. Compute the distance of the pixel (x,y) from the center (Xc,Yc)

2c

2c yyxxDis )()( -+-=

4. )((),( DisExp12yxWeight ´+= a

5. x= x + 1

6. if x <=Wn-1 goto 3

7. y=y+1

8. if y <= Hn-1 goto 2

End

2. Compute the traditional moments.

The general moment of order (p+q) that presented by Equation (2.11) can be either Centralized or Non Centralized. The Equation (2.11) can be written as follows:

),( yxfyxm qn

x y

pnpq åå=

where nn yx , are the corresponding normalized values for the coordinates x and y respectively, which can be determined by using one of the two sets of normalization equations (2.19a,b) and (2.21a,b).


۸۸

While for non centralized moment the normalized coordinates nn yx , be determined by using the following two equations:

1-=

Wnxxn ,

1-=

Hnyyn ,…….…….. (3.2)

Algorithm (3.16) Determination of General Centralize Moment

Input: Cp(x, y) array, the central part of Palmprint image,

Weight(x, y) array, Wn, Hn, p, q, Xc, Yc.

Output: pqm (general moment of order (p+q)).

Begin 1. set y=0 ,m=0 2. yy= (( y-Yc ) / Yc) ^ q 3. x=0 4. If palm shift mask=1 then mpq =mpq+ ((x-Xc)/Xc )^ p* yy* Cp(x,y)* Weight(x,y) 5. x = x + 1 6. if x <= Wn - 1 then goto 4 7. y = y + 1 8. if y <= Hn - 1 then goto 2

End where the Xc, Yc can be taken as either Xc =Wn/2 and Yc= Hn /2,

or as center of mass that described in section (3.3.9).

In non centralize moment, the same Algorithm (3.16) is followed but with the changes in steps 2, 4 as follows:

2. yy =( y / (Hn - 1)) ^ q

4. mpq = mpq + (x / (Wn - 1))^ p* yy*Cp(x,y)*Weight(x,y)

The algorithm (3.16) was used to get the values of (m00, m10,

m01, m11, m02, m20, m03, m30), which are used to compute the central moment.


۸۹

3. Compute the Central Moments

By using the general moments values (m00, m10, m01, m11, m02, m20, m03, m30), and the coordinates of the center of mass point:

00

10

mmx = ,

00

01

mm

y = ,……………. (3.3)

The central moments (M00, M20, M02, M11, M30, M03 M12, M21) up to order 3 where determined by using the equations (2.14a), (2.14b), (2.14c), (2.14d), (2.14e), (2.14f), (2.14g), (2.14h), (2.14i).

4. Determine to Normalized Central Moments.

By using the Equations. (2.15),(2.16), and (2.17) we will get the seven moments invariant.

The seven invariant moments were used as features that describe different central palmprint images.


۹۰

3.4.2 Feature Extraction Using Complex Moments

As presented in section (2.10) the complex moments were used to extract features from the central palmprint region. The first four complex moments are expressed by the equations (2.23), (2.24), (2.25) and (2.26).

The complex moment has two parts (real and imaginary), they used to compute the value of complex moment as follows:

2i

2rn MMM += )........(,......... 43

The normalized coordinates, which can be determined by using one of the two sets presented by equations (2.19a), (2.19b), (2.21a), and (2.21b).

We use the center of mass center coordinates ),( yx , and the weight for that, we must divide the result Mn by the sum of central palmprint image pixels and sum of the weight palmprint pixels, as follows:

åå åå´

+=

),(),( yxWeightyxfMM

M2i

2r

n ………..(3.5)

The four complex moment (M1, M2, M3, M4) were used as features that represent the central palmprint image.


۹۱

3.5 Enrollment Phase The enrollment process means build a database file. The enrollment phase is responsible to enroll the individual to the system by capturing hand image and extract features, a number of hand images are used to create the template of each enroll individual and give him\her identification key.

Three files needs to construct:

1. Construct the database file of individual features records.

2. Construct the mean file that contains the average of the recorded features to each individual.

3. Contrast the threshold file that contains the threshold record to each individual.

3.5.1 Database file This file consists of a number of features records extracted to each

known individual, the data of this file used to construct both mean and threshold file.

Algorithm (3.17) Data Storage Input: individual hand images, m number of images used to

create template for each individual, n number of individuals

Output: Database file consist of the extracted features record to each individual.

Begin 1. Open or create the database file to store the

extracted features records. 2. For all n individuals. 3. For each of its m. images get the features

record, and give him\her identification key. 4. Store the extracted features record in the file,

and its Id key. 5. goto 3 until no more individuals. 6. Close file.

End.


۹۲

3.5.2 Mean File

Construct the mean file that consists of the average features record to each individual this record consists of all average values of each feature beside to the identification key.

Algorithm (3.18) Mean File

Input: Database file, m number of features records to each individual, n number of individuals

Output: Mean file consists of Mean features record to each individual.

Begin

1. Open Mean file to store the average features record.

For all n individual in the database file.

2. Get the m features records from the database file that belong to the considered individual.

3. Compute the mean value to each feature listed in these m records.

4. Store the mean features record (in addition to identification key) in the mean file.

5. Goto 3 until no more individuals.

6. Close files.

End.


۹۳

3.5.3 Threshold File

Construct the threshold file that consists of threshold record to each individual, each record consists of the threshold value max allowed distance between the mean features values and the corresponding features values in any of its individual features record.

Algorithm (3.19) Threshold File

Input: Database file, Mean file, n (number of individuals), m (number of recorded features to each individual),

Output: Threshold file consists of threshold record to each individual.

Begin

1. Open Mean file and Database file.

For all n individual in the database file.

2. Get the m features records from the database file that belong to the considered individual.

For all m features records belong to that individual

3. Compute the Euclidean distance between the mean features record and individual features record.

4. Store max distance as threshold value in the record with identification key in the threshold file.

5. Goto 2 until no more individuals.

6. Close files.

End.


۹٤

3.6 Verification Phase A palmprint verification system is a one-to-one matching process.

It matches a person’s claimed palmprint to the enrolled palmprint pattern that registered. The system operation phases: enrollment and verification. Both phases consist of two sub-stages: preprocessing for palmprint localization, enhancement and feature extraction for moment features extraction. However, verification phase has an additional stage: classification (for calculating dissimilarity between the tested palmprint with the template palmprints).

At the enrollment stage, a set of the template palmprint images are processed to extract the moment features, the extracted set of moments is labeled and stored into a database. At the verification stage, from the input image a set of moment features is extracted, and then matched with the claimant’s palmprint image(s), stored in the database to get the dissimilarity degree (this was done by computing Euclidean distance metric).

Finally, the dissimilarity measure is compared to a predefined threshold to determine whether a claimant should be accepted. If the dissimilarity measure below the predefined threshold value, the palmprint input is verified and accepted as same identity as the claimed identity.


۹٥

Algorithm (3.20) Verification Phase

Input: Mean file, Threshold file.

Output: Accept /Reject.

Begin

1. After capturing hand image, preprocessing, extract features then store the features in features record with identification key input from the individual.

2. Open Mean file and threshold file.

3. Get the mean features record, threshold features record that belong to the considered individual depend on the input identification key.

4. Compute the Euclidean distance between the mean features record and individual features record.

5. If distance below the threshold value in the threshold record

then accept the individual

else reject the individual

End.


۹٦

Threshold

Accept /

Reject

Pattern matching

Features

Feature Extraction

Preprocessing

User ID

Template Stored in Data Base File

Features

Feature Extraction

Preprocessing

Enrollment Phase

Authentication Phase

Figure (3.17) Block diagram of the designed Palmprint Authentication System

Chapter Four Experimental Result and Discussion

۹۷

Person

Name ID Image No. Non Centralize Moment Invariant Features vector

1f 2f 3f 4f 5f 6f 7f

A 1

1 4.170717E-08 5.355572E-19 4.650217E-24 9.400014E-25 0 6.26419E-34 0 2 4.135055E-08 1.000657E-17 5.499654E-24 6.888533E-25 0 2.88664E-34 0 3 4.547143E-08 2.728011E-18 5.381643E-24 1.253600E-24 0 -1.54422E-33 0 4 4.611286E-08 1.621811E-17 4.320106E-24 1.131914E-24 0 -1.95772E-33 0 5 4.385452E-08 3.678343E-19 4.489299E-24 1.709505E-24 0 -3.18460E-34 0

B 2

1 9.057744E-08 1.651677E-16 9.619044E-24 8.496205E-24 0 -1.01823E-32 0 2 8.656174E-08 7.441664E-18 2.695753E-24 8.510394E-24 0 2.30948E-32 0 3 8.477332E-08 2.511841E-17 8.619463E-24 8.914280E-24 0 -2.83739E-32 0 4 6.983564E-08 1.066654E-17 4.028916E-24 6.148742E-24 0 3.57552E-33 0 5 7.977150E-08 6.844949E-17 3.613614E-24 4.108420E-24 0 -1.55533E-32 0

C 3

1 5.297428E-08 3.098834E-17 1.850451E-23 4.090800E-24 0 2.25934E-32 0 2 5.493398E-08 2.257995E-17 1.707063E-23 3.368992E-24 0 1.59654E-32 0 3 5.466323E-08 2.137613E-17 1.552882E-23 4.024646E-24 0 1.70540E-32 0 4 5.718898E-08 3.954615E-18 1.110496E-23 4.296355E-24 0 7.24819E-33 0 5 5.287964E-08 7.355694E-18 1.152687E-23 3.412172E-24 0 8.94577E-33 0

D 4


E 5


Table (4.1) The extracted non-centralized moment invariant


۹۸

Person name ID

Mean vector non-centralized moment invariant Mean 1f Mean 2f Mean 3f Mean 4f Mean 5f Mean 6f Mean 7f

A 1 4.369931E-08 5.971217E-18 4.868184E-24 1.144775E-24 0 -5.810642E-34 0 B 2 8.230393E-08 5.536875E-17 5.715358E-24 7.235608E-24 0 -5.487832E-33 0 C 3 5.452802E-08 1.725095E-17 1.474716E-23 3.838593E-24 0 1.436135E-32 0 D 4 3.762521E-08 3.126854E-17 1.980810E-24 4.490629E-25 0 1.359843E-33 0 E 5 4.985781E-08 1.991394E-17 1.299175E-23 2.943410E-24 0 8.331450E-33 0

Person name ID

Max Distance between Mean features vector and Image features vector

Threshold A 1 2.413554E-09 B 2 1.246828E-08 C 3 2.660954E-09 D 4 2.534133E-09 E 5 1.081119E-09

Table (4.3) Non-centralized moment invariant Threshold value

Table (4.2) Mean vector of Non-centralized moment invariant


۹۹

Person name ID Image No.

Feature vectors for Centralized moment invariant

1f 2f 3f 4f 5f 6f 7f

A 1

1 1.637747E-07 2.417612E-16 2.545154E-22 4.572193E-23 -2.802597E-45 6.104504E-32 0 2 1.644500E-07 3.137203E-16 3.323940E-22 3.920411E-23 -1.401298E-45 1.011568E-31 -1.401298E-45 3 1.803507E-07 2.707452E-16 3.149621E-22 6.974236E-23 -7.006492E-45 -3.362994E-31 0 4 1.865258E-07 5.274130E-16 2.663010E-22 6.404815E-23 7.006492E-45 -7.746687E-31 -1.401298E-45 5 1.713930E-07 4.088551E-20 2.745033E-22 1.053514E-22 1.401298E-45 1.523977E-32 -2.802597E-45

B 2

1 4.082249E-07 2.922719E-15 9.113372E-22 7.995693E-22 4.540207E-43 -3.106707E-30 1.751623E-43 2 3.820021E-07 4.302969E-16 2.654436E-22 7.732338E-22 2.872662E-43 1.574333E-29 2.732532E-43 3 3.787488E-07 3.284878E-16 8.204254E-22 8.496218E-22 7.090570E-43 -5.352138E-30 4.344025E-44 4 3.101563E-07 5.453208E-17 3.739430E-22 5.501302E-22 -1.611493E-43 1.534653E-30 1.429324E-43 5 3.794042E-07 1.728966E-15 3.811299E-22 4.372548E-22 -1.821688E-44 -7.994247E-30 4.624285E-44

C 3

1 2.179247E-07 1.660673E-15 1.406056E-21 3.833202E-22 2.802597E-43 1.550572E-29 1.681558E-44 2 2.248021E-07 1.348950E-15 1.270646E-21 3.109210E-22 1.933792E-43 1.141502E-29 4.203895E-45 3 2.304612E-07 1.620951E-15 1.325281E-21 4.272387E-22 3.194960E-43 1.716636E-29 4.063766E-44 4 2.423140E-07 1.157214E-15 1.024610E-21 4.723208E-22 3.279038E-43 1.606699E-29 4.484155E-44 5 2.173456E-07 8.078146E-16 9.036689E-22 3.162341E-22 1.681558E-43 8.983672E-30 1.821688E-44

D 4

1 1.540746E-07 9.091117E-16 1.814076E-22 2.445980E-23 1.401298E-45 5.871428E-31 -1.401298E-45 2 1.800800E-07 1.718369E-15 2.714488E-22 5.057561E-23 5.605194E-45 2.088702E-30 2.802597E-45 3 1.476867E-07 5.993421E-16 1.535042E-22 2.744143E-23 1.401298E-45 6.682214E-33 1.401298E-45 4 1.557495E-07 9.322260E-16 1.315865E-22 3.208992E-23 1.401298E-45 6.972851E-31 1.401298E-45 5 1.555162E-07 4.334277E-17 2.974934E-23 4.700459E-23 1.401298E-45 2.441592E-31 1.401298E-45

E 5

1 1.951116E-07 5.047873E-16 7.074929E-22 1.950940E-22 2.382207E-44 2.664640E-30 -1.961818E-44 2 2.001172E-07 1.397508E-15 1.106337E-21 3.688210E-22 1.541428E-43 1.262169E-29 -2.942727E-44 3 1.994302E-07 6.078252E-16 9.077946E-22 1.709685E-22 1.261169E-44 1.737189E-30 -2.242078E-44 4 1.988089E-07 5.770621E-16 8.183435E-22 2.005564E-22 2.942727E-44 3.352987E-30 -2.101948E-44 5 1.976314E-07 6.395747E-16 7.852095E-22 1.487749E-22 5.605194E-45 1.681317E-30 -1.821688E-44

Table (4.4) Centralized moment invariant feature vector


۱۰۰

Person name ID

Centralize Moment Invariant Mean features vector Mean 1f Mean 2f Mean 3f Mean 4f Mean 5f Mean 6f Mean 7f

A 1 1.732988E-07 2.707361E-16 2.885352E-22 6.481360E-23 0 -1.867053E-31 -1.401298E-45 B 2 3.717073E-07 1.093000E-15 5.504558E-22 6.819619E-22 2.53635E-43 1.649783E-31 1.359260E-43 C 3 2.265695E-07 1.319121E-15 1.186052E-21 3.820070E-22 2.57839E-43 1.382755E-29 2.522337E-44 D 4 1.586214E-07 8.404783E-16 1.535393E-22 3.631427E-23 2.80260E-45 7.247943E-31 1.401298E-45 E 5 1.982199E-07 7.453515E-16 8.650356E-22 2.168429E-22 4.48416E-44 4.411564E-30 -2.242078E-44

Person name ID

Max Distance between Mean features vector and Image features

vector Threshold

A 1 1.322699E-08 B 2 6.155094E-08 C 3 1.574450E-08 D 4 2.145856E-08 E 5 3.108312E-09

Table (4.5) Mean vector of Centralized moment invariant

Table (4.6) Threshold values for Centralized moment invariant


۱۰۱

Person name ID Image No.

Complex Moment Features vector

1M 2M 3M 4M

A 1

1 4.471346E-06 1.127580E-06 1.644780E-06 2.125386E-07 2 5.167641E-06 1.232742E-06 2.007788E-06 4.118983E-07 3 3.966803E-06 1.104372E-06 1.593505E-06 3.951383E-07 4 4.007393E-06 1.366570E-06 1.767649E-06 3.289326E-07 5 5.320149E-06 3.092501E-07 1.421358E-06 4.207488E-07

B 2


C 3


D 4


E 5


Table (4.7) Complex Moment value for test samples


۱۰۲

Person name ID Complex Moment Mean features vector

Mean 1M Mean 2M Mean 3M Mean 4M A 1 4.586666E-06 1.028103E-06 1.687016E-06 3.538513E-07 B 2 5.669727E-06 1.420672E-06 1.466332E-06 2.907381E-07 C 3 3.310617E-06 2.086056E-06 1.619047E-06 4.850414E-07 D 4 3.454881E-06 1.585081E-06 7.024085E-07 3.139779E-07 E 5 6.230440E-06 1.983354E-06 1.780791E-06 7.677057E-07

Person name ID

Max Distance between Mean features vector and Image features vector

Threshold A 1 1.062918E-06 B 2 1.678689E-06 C 3 5.830710E-07 D 4 1.806379E-06 E 5 1.080148E-06

Table (4.8) Complex Moment Mean vector

Table (4.9) Complex Moment Threshold values


۱۰٥


4.1 Introduction

The Palmprint authentication system was tested by using palmprint images for five persons of different ages. Five images for each person, were taken to extract palmprint features of each person.

4.2 Enrollment Phase

In the enrollment phase, as presented in section (3.5), the samples images for the five persons were enrolled by the system, with five image to each person, then the extracted features were stored in the database file, the average vector of the five sets of the extracted features (one vector for each person) were calculated and stored in the mean file with identification key for each person, finally the threshold value was computed as the max distance between the determined mean features vector and the extracted feature vectors, then the threshold value was registered with the identification key in the threshold file. All these operations are repeated on all enrolled samples.

The tables (4.1), (4.4), and (4.7), tabulates the set of extracted features using centralized moments invariant, non-centralized moments invariant and complex moments. The tables (4.2), (4.3), (4.5), (4.6), (4.8), and (4.9) show the mean and threshold values to each moment type.


۱۰٦

4.3 Verification (Authentication) Phase To evaluate the verification performance, the tests False

Acceptance Rate (FAR) and a False Rejection Rate (FRR) were performed. The false Acceptance means a biometric system incorrectly verifies an imposter against a claimed identity, while the false rejection means a biometric system will fail to identify an enrollee. These measurements defined as follows:

% 100accesses imposter of number total

claims imposter accepted of numberFAR ´= ,........ (4.1)

% 100 accesses cliant of number total

claims clieint rejected of numberFRR ´= ,........... (4.2)

The tables (4.10), (4.11) and (4.12) show the results of matching:

MR: Matching Result.

S: Client successful access to the system.

F: The imposter fails access to the system.

FA: The system accepts imposter to access the system.

FR: The system rejects client to access the system.


۱۰۷

In the matching process two new images were used to each person to verify the identity of the enrolled individual. The results of the three used moments are as follows:

Moment Type FA FR F S Centralize moment invariant 3 1 37 9 Non Centralize moment invariant 0 1 40 9 Complex moment 9 2 31 8

By computing the FAR and FRR the performance of the non-

centralized moments is the best among others.

Moment type FAR FRR Centralize moment invariant 0.075 0.1 Non Centralize moment invariant 0 0.1 Complex moment 0.225 0.2

۱۰۸

Chapter Five Conclusions and Future Work Suggestions

This chapter is dedicated to present some derived conclusions, extracted from the investigations to the test results. Also, some suggestions for future extended work are presented.

5.1 Discussion and Conclusions In this research work a new palmprint authentication system is

proposed. The digital scanner was used to capture the hand images. A preprocessing step is used to segment the central palmprint area from other areas (like fingers and background), then the brightness-contrast enhancement is applied on the palmprint image to improve the extraction of the principle lines. Sobel operator is used to detect of the edges in the palm.

Moment invariant and Complex moments are used to extract the palmprint features, then these features are used for matching (using Euclidean distance as a similarity measure).

To evaluate the verification performance the False Acceptance Rate and False Rejection Rate were determined for all cases of moments. It is found that the non-centralized moment invariant shows the better performance than the other moments, it's FAR=0 and FRR=1%.

5.2 Recommendations The following recommendations are suggested to develop the

system performance:

1. Using another type of image enhancement to improve the appearance of the lines in the palm, we can use the histogram equalization.

2. Using other methods of edge detection to improve the detection of principal lines, such as using laplacian filter.

۱۰۹

3. Using other types of moments in order to extract the palmprint features, like using orthogonal moments (Zernike and Legendre).

[1] Garcia J. O., Bigun J., Reynolds D., and Rodriguez J.G, "Authentication Gets Personal with Biometrics", IEEE SIGNAL PROCESSING MAGAZINE, MARCH 2004

[2] Biometric Identification versus Verification http://www.findbiometrics.com/Pages/guide4.html

[3] FUJITSU,"First Contact less Palm Vein Pattern Biometric Authentication System" Laboratories Develops Technology for Word's, 2004.

http://www.FUJITSU.com/global/ [4] Biometric Glossary http://www.biometricwatch.com/glossary.htm [5] Spence, B., "Biometrics' Role in Physical Access Control",

Recognition System Hand Geometry Biometric http://www.handreader.com/news/casestudies/cs25.htm [6] What are Biometric http://www.isl-biometrics.com/solutions/biometrics.htm

[7] JCSIG introduction to Biometric, Java Card Special Interest Group http://www.javacard.org/others/biometrics_intro.htm [8] Orla O'Sullivan, "Biometric come to life". http://www.banking.com/aba/cover_0197.htm [9] Teach Overviews http://www.biometricwatch.com/tech_overview.htm

[10] Zhang, L. and Zhang, D.,"Characterization of Palmprints by Wavelet Signatures via Directional Context Modeling",

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, VOL. 34, NO. 3, JUNE 2004.

E-mail: [email protected] [11] "Characteristic of successful Biometric Identification Methods" http://www.technovelgy.com/ct/TechnologyArticle.asp?ArtNum=11 [12] Doi, J. and Yamanaka, M.,"Personal Authentication Using

Feature Points on Finger and Palmar Creases", Chiba Institute of Technology, Department of Computer Science, Proceedings of the 32nd Applied Imagery Pattern Recognition Workshop© 2003 IEEE.

E-mail:[email protected], [email protected] [13] You, J., Kong, W.K., Zhang, D., and Cheung, K.H.," On

Hierarchical Palmprint Coding With Multiple Features for Personal Identification in Large Databases", IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 14, NO. 2, FEBRUARY 2004.

E-mail: [email protected] [14] Kong, W. K., and Zahnag, D.,"Palmprint Texture Analysis based

on Low-Resolution Images for Personal Authentication.", Biometric Research Center, Department of Computing .The Hong Kong Polytechnic University, Kowloon,HongKong, Published by the IEEE.

E-mail: [email protected] E-mail: [email protected] [15] Zhang, D., Kong W.K., You J., and Wong, M.,"Online Palmprint

Identification.", IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 25, NO. 9, SEPTEMBER 2003, IEEE Published by the IEEE Computer Society Recommended for acceptance by K. Yamamoto.E-mail: {csdzhang, cswkkong, csyjia, csmkwong}@comp.polyu.edu.hk.

[16] W. W. Boles and S. Y. T. Chu,"Personal Identification Using

Images of the Human Palm", Signal Processing Research Centre School of Electrical and Electronic Systems Engineering Queensland University of Technology, [email protected], 1997 IEEE TENCON - Speech and Image Technologies for Computing and Telecommunications.

[17] WU, X., WANG, K., ZAHNAG, D.,"WAVELET BASED PALMPRINT RECOGNITION", Proceedings of the First International Conference on Machine Learning and Cybernetics, Beijing, 4-5November 2002. EMAIL:[email protected],[email protected],[email protected].

[18] Umbaugh, S.E., "Computer Vision and Image Processing", A

Practical Approach Using CVIP Tools", Prentice-Hall, Inc., 1998 [19] Gonzalez, R.C., and Wintz, P.,"Digital Image processing", second

edition, Addison-Wesley Publishing Company, 1987.

[20] "The BMP Image file format" http://astronomy.swin.edu.au/~pbourke/dataformats/bmp/ [21] Bourke, P., "BMP image format", July 1998 http://www.prepressure.com/formats/bmp/fileformat.htm

[22] "Displaying BMP file format"

http://www.funducode.com/freec/Fileformats/format1/format1.htm

[23] L. Leurs." Bitmap versus vector graphics".2000 http://www.prepressure.com/image/bitmapvector.htm

[24] Bazi, Y.,"A Comparative Study of Histogram Based Thresholding Algorithms"

http://science.unitn.it/~tomasi/think/pdf/abazi.pdf

[25] "HISTOGRAM THRESHOLDING" http://www.s2.chalmers.se/undergraduate/courses/ess060/PDFdocuments/ForPrinter/Notes/Thresholding.pdf

[26] Haykin, S.,"Intelligent Image Processing". Steve Mann

Copyright.2002 John Wiley & Sons, Inc. [27] "Pattern Recognition Principle", U.S.T.O., Electronic Institute,

Vision and Pattern Recognition. http://www.angelfire.com/linux/asthussain/MPMSC1.pdf [28] SKLANSKY, J.,"Pattern Recognition Introduction and

Foundations", University of California at Irvine, Hutchison&Ross, Inc., 1973.

[29] "Introduction to Pattern Recognition" October 15, 2004 http://bruce.engr.ucf.edu/~mge/courses/eel5825_fall04/dtt(part1).pdf

[30] Dr.s.Bell Kasim,"Moment descriptors"

http://watfor.uwaterloo.ca/mkamel/SD776/NOTES/note2.pdf

[31] Gerhard X. Ritter; Joseph N. Wilson," Handbook of Computer Vision Algorithms in Image Algebra", ISBN: 1996

[32] "Moment Invariant Features"

http://utopia.poly.edu/~ademir02/science/11/11x.html

الخالصة

نظام التحقق من هوية الشخص باستخدام راحة اليد

يهـــــدف البحـــــث اىل بنـــــاء منظومـــــة للتحقـــــق مـــــن هويـــــة الشـــــخص باســـــتخدام راحـــــة

ـــات التشـــببات الـــيت تتســـ بـــا راحـــة اليـــد مـــن الســـمات الـــيت متيـــز اليـــدححيث تبتـــ الترـع

ــد اد ا اههتمـام بــذا املوضــوع خـالل الســنوات امل اضـية اـتبــار ذلــك ــخص ـــن اهخـع ـ

سيلة حيوية للتمييز التحقق من اهلوية.

تبت بصمة اليد من الصرات احليوية اجلديدة نسبيا الـيت تسـتخدم للتحقـق مـن هويـة

الشـخص كو�ــا تسـتثمع الصــرات املميـزة لعاحــة اليـد ( منهــا اخلطـوط اهساســية ح التشــببات ح

بانــه مييـــز ـــخص ــــن اهخعحيــث ان هـــذه التجاـيــد القمــ ). ان نســـيج راحــة اليـــد يتســ

الطعيقـة تتميـز با�ــا مقبولـة مــن ـبـل اهـــخاصح ـليلـة الكلرــة حيـث ان حتديــد صـرات نســيج

اليد ه حتتا اىل ههاد ـا الدـة كما بصمة اهبام.

ـة مـن ا استحصال صورة اليـد باسـتخدام املاسـا العـو مث ا اخعـبا الصـورة لمـو

ميــة املمكنــة حــىت ا احلصــول ـلــى منطقــة راحــة اليــد فقــطح ان تقنيــات املباجلــة ال صــورية الـع

ميـــة تـــؤثع ـلـــى كرـــاءة ـمليـــة التحقـــق لـــذلك فـــان املباجلـــة الصـــورية ــــة املباجلـــة الصـــورية الـع

ـد اسـتخدما مية همث معحلة ضع رية ـبل معحلة اسـتخعا خصـا ص بصـمة راحـة اليـد ـ الـع

د.البز م هستخعا خصا ص راحة الي

ان اي نظـــام حتقـــق يتعـــمن مـــعحلتني اساســـية معحلـــة التســـجيل هـــمث معحلـــة ا خـــال

الرــع اىل النظــام الــيت تتطلــب اخــذ ـــد مــن النســخ املختلرــة لليــد اـطــاءه مرتــاح تبعيرــمث

ـة مـن ـمليــات املباجلـة الصــورية ابتـداءا مــن للـدخول اىل النظـامح اخعــبا هـذه النســخ لمـو

عض اليها الصورة ـنـد ـمليـة اه خـالح اسـتقطاع املنطقـة املعكزيـة لعاحـة ادالة العوضاء اليت تتب

ــــة ــــد باســــتخدام تقني ــــة ابــــعاد اخلطــــوط راحــــة الي اليــــد باســــتببا اهصــــابع اخللريــــة ا حما ل

احلواف. ا استخعا اخلصا ص من خالل احتساب البز م هلاح ا ا خزن مبدل البز م لكل

خدامه ـند ـملية التحقق كـذلك ا خـزن ـيمـة البتبـة الـيت تـت فع مع مرتاحه التبعيرمث هست

ـلى اساسها ـملية رفض الرع ا ـبوله ـد ا احتساب هذه القيمة مـن خـالل اـلـى ـيمـة

فعق (مقياس للتشابه) ـن املبدل.

اما املعحلة الثانية همث معحلة التحقق اليت تت ـند خول الرع اىل النظام باخـذ صـورة

ــة املباجلــات الصــورية ليــده ا خــال املرتــاح التبعيرــمث مــن ـبلــهح اخعــبا الصــورة لــنرس جممـو

ـند معحلة التسجيل ا ا احتساب البز م هلا حساب مقدار التشابه مـع النسـخة املخز نـة

لــه مبقارنــة هــذه القيمــة مــع ـيمــة البتبــة املخز نـــة لــه ح فــاذا اـــل فهــو مقبــول اذا اكثــع فهـــو

معفوض.

يهـا املعكزيـة الغـري هذا ان من البز م ح ا ه البز م الثابتـة بنـو البحث ا جتعبة نـو

معكزيـــةح ثانيـــا البــــز م املبقـــدة. لقيــــاس ا اء املنظومـــة اســــتخدم مقياســـان مهــــا مبـــدل الــــعفض

اخلـاطئ لــدخول الشـخص املخــول اىل النظــام مبـدل القبــول اخلــاطئ لـدخول الشــخص الغــري

حملتال).خمول اىل النظام (ا

اظهــعت النتــا ج ان البــز م الثابــة الغــري معكزيــة هــمث افعــل أ اءا ـمليــة التحقــق مــن

هويــــة الشــــخص (لبــــد مخســــة اـــــخاص) حيــــث ان ـــــد مــــعات الــــعفض اخلــــاطئ للشــــخص

املخول كانا (صرع) ـد معات القبول اخلاطئ للشخص احملتال كانا ( احد) .

ÉÑÇÒæ العلمي والبحث العالي التعليم بغداد جامعة العلوم كلية الحاسبات علوم قسم

نظام التحقق من هوية الشخص باستخدام راحة اليد

أطروحة

بغداد جامعة يف العلوم كلية اىل مقدمة احلاسبات علوم يف املاجستري درجة نيل متطلبات من كجزء

قبل من

مها محمدعلي شريف الطريحي )٢٠٠٢الحاسبات علوم بكالوريوس(

بغداد جامعة

اشراف

Ï.لؤي ادور جورج

æÐ االول كانون احلجة

2004 1425

hand palmprint authentication system - scbaghdad.edu.iq science/ms.c/2004/hand... · ministry of...

Documents