Facial recognition is a crucial factor of everyday identification processes: human beings recognize and evaluate each other by means of the face. Whenever driving licences, identity and membership cards are checked or wherever access is controlled by security staff, the identity is verified by looking into somebody's face. Thus, unlike other biometric features, e.g. the fingerprint or iris recognition, facial recognition is a transparent procedure well-known to human beings. However, especially in the context of the international fight against terrorism it has become obvious that the traditional way of identifying individuals is insufficient. There are certain limits to the natural recognition process carried out by human beings: The recognition performance is not only impaired by difficulties with the recognition of people from other ethnic origin or deceptions due to a different hair-do or beards, but also by subjective impression based on a person's outward. The requirement of successful personal identification in access control and in other cases leads to using the results of biometrics. Biometrics Face recognition is a passive, non-invasive method for verifying the identity of a person, Offers the benefits of its unique facial technology in the form of customized overall solutions for the areas of access control, border control, ID-Management, search for criminals and video surveillance Face recognition has come to be an active research area with numerous applications in recent years. In this thesis, a variety of approaches for face recognition are reviewed first. These approaches are classified according to basic tasks i-e Face Detect, face Normalization, and Face recognition. Then, an implementation of the face recognition method, the Eigenface recognition approach is presented in detail as well as other face recognitions methods i-e Local Feature Analysis, Neural Networks and Automatic face processing are discussed in general.
Ever since the birth of first mankind, human beings have continually been seeking for personal possessions. From the very basics of food and clothes, to cars, houses, and the more recent substantial property of data and information, it is becoming increasingly important that such valuable assets be sheltered by means of security control. Throughout history, the types of technologies used on the access control systems are countless. From the traditional systems such as security guards checking personal ID's to the very fundamentals of keypads and locks and password or entry code, the focus now has moved to the more advance technologies, particularly in today's multifaceted society. Organisations are continuously seeking for a more secure, suitable and economical way of property protection. The problem associated with traditional mechanisms is that the possessions could be lost, stolen, forgotten, or misplaced. Furthermore, once in control of the identifying possession, any other "unauthorised" person could abuse the privileges of the authorised user. Therefore there is a need of another approach to properly differentiate the correct (right) person from an impostor by positive identification of the person seeking access. Biometrics is one rising development in the field of access control system that provides true identification. Although the word ``biometrics'' sound very new and high tech, it is in fact the oldest form of identification known to man. Since the dawn of man, a persons face and voice was used to identify him/her. Before the digital age, a hand written signature was the only method used by a person to assert a unique form of identification that was difficult to copy. Popular biometric systems in use today include fingerprint recognition, iris recognition, voice recognition, and facial recognition systems. These systems are in practice in different organizations like banks, airports, social services offices, blood banks and other highly sensitive organizations. Biometric system offers the most accurate authentication solution and convenience. Biometrics systems can be integrated into any application that requires security, access control, and identification or verification of people. With biometric security, we can dispense with the key, the password, the PIN code; the access-enabler is human beings - not something he/she know, or something in his/her possession.
Chapter 02 This part of the dissertation provides the general overview of biometrics. Definitions such as 'Automatic', 'Physiological' and 'Behavioural' characteristics are also discussed as well as different types of biometric systems i.e. one-to-one and one-to-many. General Biometrics Base systems model, how it works and Multimodal Biometrics systems are also discussed in detail. In the last section of this chapter, a comprehensive overview of the right approach in selection of different technologies for an origination in terms of business objective, user acceptance, FFR, FAR, organisational environments, cost and a comparison of all biometrics are also presented.
Chapter 03 Different types of biometric technologies are described in this chapter i.e. finger prints, iris and retina, voice, biometric signature and how these technologies work and the main features of these technologies with the help of diagrams.
Chapter 04 This chapter is one of the most important chapters which explain the general back ground of face recognition technology and how face recognition works. It gives a brief discussion of how verification and identification is achieved with the help of face recognition system. Actual techniques involved during face verification and identification i.e. faces detection, face normalisation and face recognition are also discussed in detail. Steps involved during the face detection i.e. coarse detection phase and refined search phase are discussed as well as how Normalisation is achieved through different steps i.e. lighting normalisation, scaling normalisation, rotation normalisation and background subtraction. Face recognition and methods of face recognition i.e. Eigenfaces, feature analysis, neural network and automatic face processing are discussed in this presentation.
Chapter 05 In this chapter of my dissertation, a proposed model of face recognition system for attendance of university students is discussed. The specification of the system is also compiled after the extensive study of face recognition products of different Vendors.
Chapter 06 This final chapter of my dissertation contains the conclusion, future work and issues involved with face recognition system.
A review of the biometrics technology
Biometrics: An overview In today's networked and digital world the role of system security has a vital importance. In originations a large number of professional people are involved in one form of electronic transaction or another. Securing a company's digital assets and identities is a necessity for financial success. Ignoring IT security increases the risk of losses for any company moving through this electronic world. Logging on to a system by entering user ID and password is very simple but its simplicity makes serious security problems. There are, however, people who use 'easy guess' passwords or leave written passwords near to their computer. In this situation there is no way to confirm that the person is logged on the system using his/her ID and password or some one else, nothing can prevent someone else from accessing sensitive material. It's like a passport system that doesn't require a photograph. In addition, time consuming tasks behind the management of user ID and passwords divert already insufficient resources from other important responsibilities. Establishing an accurate identity is the main focus of the information systems security in recent years and great efforts are made in this field. Two types of identification systems are in use now today.
- In one type identification system flawed identity checking results in unnecessary duplication, fraud and client disruption, resulting costs and risks.
- While in other type of identification system an accurate identification procedure and effectiveness may be undermined by unpopularity resulting falsification and evasion.
Chapter2 Three conventional forms of identification are in use.
Biometrics is a branch of science in which we study, what makes us biologically unique. It is also referred to the science and application of statistical analysis of biological characteristics (Physiological/ Behavioural). In security terms, Biometrics refers to technologies that analyse human characteristics for security purposes. Therefore Biometrics technologies are concerned with the physical parts of the human or personal trait of human being. There are different definitions of security base biometrics that have been circulating for a numbers of years. According to Ashbourn, an expert in Biometrics, "Biometrics is a measurable physiological and / or behavioural trait that can be captured and subsequently compared with another instance at the time of verification").  The Biometrics Consortium states "Biometrics is automated methods of recognizing a person based on a physiological or behavioural characteristic".  The international Biometrics Group defines biometrics as "the automated use of physiological or behavioural characteristics to determine or verify identity" 
- Origination ID or smart cards.
- The use of passwords or Personal Identification Number's, mother name, place of birth, home address etc.
- The third form of identification is to identify something unique about a person, such as fingerprints, voice recognition, hand geometry, face structure, iris and retina. This third form of identification is known as 'Biometrics'.
As mentioned, biometric technologies are anxious with the physical parts of the human or personal mannerism of human beings. The word "automatics" basically means that biometrics technology must recognise to identify /verify human characteristics rapidly and automatically, in real time. Unique physiological characteristics or behavioural mannerisms are examined in biometrics verification for an individual's identity. Physiological characteristics are essentially unchangeable such as hand geometry, iris pattern , palm prints, face structure and vane shape etc .while behavioural characteristic such as one's signature, voice or keystroke dynamics are changeable and these behavioural characteristics can change over time. They are both controllable and less controllable actions. The initial sample of the biometrics template, which is stored in the data base during the Enrolment, must be updated each time it is used. Although behaviour characteristics based biometrics is less costly and less intimidating to users, physiological characteristics have a tendency to offer greater accuracy and security. In any case, both techniques grant an extensively higher level of identification and verification as compare to smart cards or passwords technologies. A password or personal identification number (PIN) is not unique for an individual ,it can be stolen ,forgotten or lost, while a biometric characteristic is unique to each individual; it can be used to prevent fraud or theft. It cannot be lost, stolen or forgotten. There already many places such as research laboratories, defence (military) installations, VIP offices, day care centres and cash points where access is guarded by biometrics base authentication system. The following biometric identifiers currently obtainable or under development are fingerprints, body aroma, ear shape, face recognition, keystroke dynamics, palm print, retinal scan, iris pattern, signature, DNA, "vein check" and voice pattern.
- Physiological characteristics are fingerprint, Hand geometry, iris pattern ,retinal, ear shape and facial scans etc
- Behavioural characteristics are voice pattern, key strokes, signature etc.
Biometrics-based Systems A biometric based system is a system that in some way uses physical characteristics or personal traits of a human being. These systems are not only, mainly used for security, but also use for encryption.
Encryption The processes of translating a message (plaintext), with the help of software, into a programmed message/encoded text (Cipher text), called Encryption. This is usually accomplished using a secret key and a cryptographic code. 
Type of Biometrics-based Systems There are two types of Biometrics-based systems. One-to-one systems (Verification system) One-to-many systems (Identification System)
One-to-one system (verification) This type of biometric system works on the base of one to one matching and authentication principles where the system asks and attempts to answer the question "Am I who I claim to be?" At first a biometric sample of a person is provided to the system and then the system matches this sample to the previously stored template during the enrolment mode for that person. The system then decides whether that is the person who claims the identity. After a successful matching of the fresh sample with the stored template, the system authenticates the person. These types of systems are also referred to as verification systems. The verification system is a fast response system because it minimises the use of resources and time by providing biometrics sample/ information to the system which specifies the stored template in the data base for that person. 
One-to-many system (identification) This type of biometrics system works on the base of one to many recognition principles. The system attempts to answer the question," Who am I?" The basic purpose of this system to identify a person's identity by performing matches against all biometrics templates stored in a data base or a data library. A person does not claim his/her identity to the system; instead the person just gives the system some biometric data. The system then performs to match this data to all templates previously stored in the database and decides whether a match can be made. It is not necessary that the system responds with the person's name, it could be the person's ID or other unique identity. These types of systems are referred to as identification systems . Identification systems have a slow response as compared to verification systems. This is because they require much more powerful resources due to the fact that more comparisons are required by identification systems. The biometrics identification system also prevents a person from registering twice on the system and ensures that a person is not already present in a data base. This type of system can be used in a large scale public benefits organisation, such as being used at banks where a person would try opening a second account on another name. This system can also be used with immigration where a person could try to enter the country on false documents.
General Biometrics Base Authentication System Model A general biometrics base authentication system model consists of three major components, hardware, software and interface. Hardware is used to capture the biometrics information and software is used to maintain and manage it while an interface with application system that will use the result to confirm an individual's identity. The system operates in two different modes:
- Enrolment mode
- Authentication mode
Enrolment mode: In this mode a user's biometrics data is provided to a system, which stores this user's biometric sample in a database or data library as a template. Hardware such as a biometrics readers/ scanners, cameras are used to capture biometrics sample. This stored template is then labelled with a user identity e.g. name, identification number etc.
The way biometrics operate Some biometric base authentication systems may need a number of biometrics samples in order to build a profile of the biometric characteristics. These exclusive characteristics are then extracted and changed in to mathematical code by the system. Which is then stored in to the biometric system as a biometric template for the person who enrolled? The template is store in the memory storage of the system, or in computer database, smart card or barcode. A threshold is set in to the biometrics base authentication system according to the level of security , (a high threshold is set for high level of security) To secure the template to the person, a trigger or other mean of securing such as personal identification number, or a smart card that store the template which read by a card reader during the authentication mode, are use in biometrics. In some biometrics system when ever a person interacts with the system a new biometrics sample is provide to the system which is compared to the template. If this new sample and stored template is match (the score of new match if exceed from the set threshold then access is granted to that person). As both physical and behavioural characteristics are inconsistent with time, this change may be due to the age of the person, general health condition, working and environmental conditions and time pressures etc. the biometric base authentication system must allow for these delicate changes, in this case before a match is recorded a threshold *1 is set. This can take the form of an accuracy score *2. The comparison between the template and new sample must exceed this set threshold. If it not exceeds the system will not record the match and will not identify the person. This use of a threshold gives biometric technologies a significant advantage over passwords, PIN's and ID badges. The use of a threshold affords a tremendous degree of flexibility and if the comparison between the new biometric sample and the template exceeds the stated threshold, identity will be confirmed.
Capture, extraction, comparison and match/non match are the four stages use by all biometric authentication systems.
- Threshold:-a predefine number, often controlled by system administer, which establish the degree of correlation necessary for a comparison to be deemed a match.
- Score: - A number indicating the degree of similarity or correlation of a biometrics match
- Capture - A physical or behavioural sample is captured by the system during enrolment.
- Extraction - unique data is extracted from the sample and a template is created.
- Comparison - the template is then compared with a new sample.
Multimodal Biometric System In some environments a signal biometrics identifier base system such as finger scan, face scan or iris scan etc often not able to meet the desired performance requirement of the organization. Different biometrics base identification system such as face recognition, finger print verification and vice verification, is integrated and worked as a single biometrics base identification system. Multimodal biometrics base identification system is use to over come the limitation of the single identifier biometrics base identification system. Initial experimental results reveal that the identity established by such an integrated system is more reliable than the identity established by a signal biometrics identifier base system. 
Selecting the Right Approach In Different Environment Different biometrics base authentication systems are used. To choose the right approach to biometrics authentication it is necessary to understand the requirement of the organisation, the application of the biometrics system, and characteristics of the biometrics devices itself. Following factors are also important to choose a biometrics base authentication system, which most devices can't store raw fingerprints and that fingerprints can't be reconstructed based on the data stored within these systems. Intrusiveness is another factor affecting user acceptance of some devices, particularly iris and retinal scanning systems. 
Business objective of the organisation The most important aspect to consider when selecting a biometrics base authentication system is the organisation business' objectives. The choice biometrics system must meet or exceed organisational business objectives as well as sustain organisation in the coming years. Business objective is the bottom line where organisation starts and end.
User acceptances Some biometrics, such as fingerprints, may be apparent as an assault of personal privacy. The system must not associate with other govt agencies biometrics (finger print) recognition system that most devices can't store raw fingerprints and that fingerprints can't be reconstructed based on the data stored within these systems. General intrusiveness can be another factor affecting user acceptance of some devices, particularly iris and retinal scanning systems. Following are the errors of biometrics base authentication system.
False acceptance rate (FAR) False acceptance rate (FAR) is a system error. It is the rate at which an interloper can be recognized as a valid user. In one -to-one match during user verification, false acceptance is based on fake attempts, not on the total number of attempts by valid users. If FAR is 1%, it means one out of 100 users trying to break into the system will be successful . FARs become more critical when you attempt to identify users based on biometrics, instead of simply trying to verify a person with a one-to-one or one-to-few operation
False reject rate (FRR) False reject rate (FRR) is another type of error of biometrics system. It is the rate at which a valid user is rejected from the system. Consider a finger print recognition system; unfortunately, the conditions under which the original sample was collected can never be exactly duplicated when the user submits subsequence biometrics information to be compared. False reject rate may occur due to following variations.
To over come FRR it is essential that all biometrics base authentication systems have a threshold value in order to allow for minor differences. With out threshold value FRR occurs and valid users will be probably rejected by system. If the threshold value is too high FAR occur . It is there for necessary to find a proper threshold value.
- Rotation and Translation because of different positioning of the finger on the finger print device.
- Downward pressure on the surface of the input device which changes the scale of input device.
- Non-permanent or semi-permanent distortions like skin disease, scars, sweat, etc
Organisational environments As stated it is important to consider the organisational environment when selecting biometrics base authentication system. Users with wet, dirty or dry hand have experienced problems with finger and palm recognition system. People using gloves generally can't use these systems. Face recognition system can't be easily be used in medical environments where hood and masks are used by users.
Cost The direct cost of the system (hardware and software) is the initial considerations. Due to the improvement of features and functionality the over all cost of biometrics system reduces. It not only reduces fraud and eliminating problems associated with stolen or forgotten passwords but also reduces the help desk role.
Summary The subject of this chapter is biometrics, which is defined as "...a method of verifying an individual's identity based on measurement of the individual's physical feature(s) or repeatable action(s) where those features and/or actions are both unique to that individual and measurable". A biometrics system which consists of enrolment mode and authentication mode, unique physiological characteristics or behavioural mannerisms are examined in biometrics verification for an individual's identity. All biometric systems essentially operate in a similar way in a four-stage process that is automated and computerized which are Capture, Extraction, Comparison and Match/non-match. Biometrics system one-to-one is based on one to one matching and authentication principles and is mainly used for verification purposes, while biometrics system one to many works on the principles of one-to-many recognition and is used for identification. Multimodal biometrics base identification system is used to over come the limitation of the signal identifier biometrics base identification system in which different biometrics base identification system such as face recognition, finger print verification and vice verification, is integrated and worked as a single biometrics base identification system.
Methodologies of Biometrics Authentication
An overview As stated, different biometric systems are use in different organisations according to their requirements. The most common biometrics system in use today includes fingerprint recognition, iris recognition, and voice recognition and face recognition systems. There are also other biometric systems available like retina recognition, vein pattern recognition, signature and DNA matching systems. These systems are not as widely used yet for various reasons. These biometrics systems can be integrated into any application that requires security, access control and identification or verification of people. With biometric security we can dispense with the key, the password and the PIN code; the access-enabler is a person, not something person know or something in his /her possession. Biometrics systems secured resources are based on who a person is. Biometrics systems also minimise the risk that is associated with less advanced technologies while at the same time offering a higher level of security and convenience.
Fingerprint Recognition System Fingerprints are one of the human physiological characteristics that do not change throughout someone's life. Even identical twins have different fingerprint patterns. The chance of identical twins to have the same fingerprint is less than one in a billion. Fingerprint recognition is generally considered the most practical system for its reliability, non-intrusive interfaces, and cost-effectiveness. In recent years, fingerprints have rallied significant support as the biometric technology that will probably be most widely used in the future. In addition to general security and access control applications, fingerprint verifiers are installed at different organisations such as, defence/military organisations health care, banking and finance, application services providers, immigration, law enforcement etc. The fingerprint's strength is its acceptance, convenience and reliability. It takes little time and effort for somebody using a fingerprint identification device to have his or her fingerprint scanned. Studies have also found that using fingerprints as an identification source is the least intrusive of all biometric techniques.  Verification of fingerprints is also fast and reliable. Users experience fewer errors in matching when they use fingerprints versus many other biometric methods. In addition, a fingerprint identification device can require very little space on a desktop or in a machine. Several companies have produced capture units smaller than a deck of cards. One of the biggest fears of fingerprint technology is the theft of fingerprints. Skeptics point out that latent or residual prints left on the glass of a fingerprint scanner may be copied. However, a good fingerprint identification device only detects live fingers and will not acknowledge fingerprint copies. 
Main Feature of Finger print verification system
- Analysis of minutia points i.e. finger image ridge (verification) endings, bifurcations or branches made by ridges.
- One of the most commercially successful biometric technologies.
- Important for applications where it is necessary to verify the identity of those who gain access.
How fingerprint recognition system works In biometrics systems fingerprint recognition system is the fastest verification /identification (One-to-One / One-to-Many) system as shown in figure 3, 4, 5. Like other biometrics recognition systems it performs fingerprint recognition with the help of specialised hardware. This specialised hardware is supported by the conventional computer hardware and special software. All biometrics systems operate in two modes, enrolment mode and authentication mode (as discussed in the previous chapter). A sample of the fingerprint of a live person is provided to the system which is then converted into mathematical code (Template) and stored for the enrolee into the database. In the first step of the authentication process, a fingerprint impression is provided to the system. The system takes a digital image (input image figure 3:1:1 below) using different techniques including scanner, optical, and ultrasound or semiconductor chip technologies. The digital image of the fingerprint includes several unique features in terms of ridge bifurcations and ridge endings, collectively referred to as minutiae.  In the next step the system uses an automatic feature extraction algorithm to locate these features in the fingerprint image, as shown in Figure 3:1:2. Each of these features is commonly represented by its location (x, y, and z) and the ridge direction at that location; however the feature extraction stage may miss some minutiae and may generate spurious minutiae due to sensor noise and other variability in the imaging process. The elasticity of the human skin also affects the feature extraction process.  In the final stage, a final decision of match and non match is made on the bases of similarity between the two sets of features after compensating for the rotation, conversion and dimension. This similarity is often expressed as a score. A decision threshold is first selected. If the score is below the threshold, the fingerprints are determined not to match; if the score is above the threshold, a correct match is declared an authentication is granted to the person.
Iris and Retina Recognition System Biometrics which analyse the intricate and unique characteristics of the eye can be divided into two different fields, Iris and Retina. Iris and retinal scans both deal with the human eye. They are done in an extremely different way as compared to other biometrics technology.
Iris Recogniton System Iris recognition biometrics base authentication systems have unique characteristics and features of the human iris used to verify the identity of an individual. The iris is the area of the eye where the pigmented or colour circle, usually brown or blue, rings the dark pupil of the eye. It consists of over 400 unique distinguishing characteristics that can be quantified and used for an individual identity. However, only about 260 of those characteristics are captured in a "live" iris identification process . Iris' are composed before birth and, except in the event of an injury to the eyeball, remain unchanged throughout an individual's lifetime. Eyeglasses and contact lenses present no problems to the quality of the image and the iris recognition /scan systems test for a live eye by checking for the normal continuous fluctuation in the pupil size. As Iris patterns are extremely complex and unique they carry an astonishing amount of information. The fact that an individual's right and left eye are different and that patterns are easy to capture, it establishes iris recognition /scan technology as one of the biometrics that is very resistant to false matching and fraud. The false acceptance rate for iris recognition systems is 1 in 1.2 million, statistically better than the average fingerprint recognition system. The real benefit is in the false-rejection rate, a measure of authenticated users who are rejected. Fingerprint scanners have a 2 percent false-rejection rate whereas iris scanning systems boast rates at a 0% level .
How Iris recognition systems work Like other biometrics base authentication systems it consists of two modes, enrolment and authentication mode. In the iris recognition/scan process a photograph of the eye is taken with the help of a specialised camera, typically very close to the subject (eye), no more than three feet. This specialised camera uses an infra-red imager to illuminate the eye and capture a very high-resolution photograph. This process takes only one to two seconds and provides the details of the iris that are mapped, recorded and stored for future matching/verification. Two types of methods are used in iris recognition process.
In Active iris identification method the distance between the user and the camera must be between six and fourteen inches. It also requires the user to move back and forth so that the camera can adjust and focus in on the user's iris. The passive system differs as such that it allows the user to be anywhere from one to three feet away from the integrated series of cameras that locate and focus in on the iris. In the identification process a wide-angle camera calculates the position of the eye. When a person stands in front of the iris identification system, between one and three feet away, a second camera zooms in on the eye and takes a black and white image. Once the iris is in focus, it overlays a circular grid on the image of the iris and identifies the light and dark areas, like an "eye print". The captured image or "eye print" is checked against a previously stored reference template in the database. The inner edge of the iris is located by an iris recognition /scan algorithm which maps the iris' dissimilar patterns and characteristics. An algorithm is a series of commands that tell a biometric system how to interpret a specific problem. Algorithms have a number of steps and are used by the biometric system to verify if a biometric sample and record is a match.
- Active iris recognition process
- Passive iris recognition process
Retina Recogniton System The retina is the layer of blood vessels at the back of the eye. Like iris recognition biometrics base authentication systems, Retina recognition /scanning are also an extremely accurate biometric system. The patterns of blood vessels at the back of the human eye are unique, and it remains the same throughout ones life.
How Retina recognition system works Retina scans are performed by directing a low-intensity infra-red light to capture the unique retina characteristics. In order for retina scan devices to read through the pupil, users must situate their eyes within three inches of the capture device and hold still while the reader finds the blood vessel patterns. They must also focus their eyes on a single point of light to obtain a successful reading. An area known as the face, situated at the centre of the retina, is scanned and the unique pattern of the blood vessels is captured. Retina biometrics is considered to be the best biometrics performers. However , despite its accuracy, this technique is often thought to be inconvenient and intrusive, hence, it is difficult to gain general acceptance by the end user. The retinal scanner requires an individual to stand still while it is reading the retinal information. Eye and retinal scanners are ineffectual with the blind and those who have cataracts. 
Main feature of Iris and Retina Recognition system
- Analysis of the iris, which is the colure ring of tissue that surrounds the pupil of the eye.
- This is a highly mature technology with a proven track record number of application areas.
- The retina is the layer of blood vessels situated at the back of the eye. The scanning technique to capture data from the retina is often thought to be the most inconvenient for end-users.
- An end user must focus on a green dot, and when this has been performed, the system uses a harmless beam of light to capture the unique retina characteristics.
Voice Recognition system Speech is the primary and most natural form of communication among humans. Because of this and the fact that speech is a primary form of personal recognition, people commonly have no problem accepting it as a biometric. Voice recognition based personal identification is one of the oldest and better-accepted biometric technologies. Voice recognition technology is mainly use for verification .it is easy to use, less expensive and non-threaten to the user. Voice recognition is the only technology that offers remote personal identification with existing resources. However, if a personal recognition system developer requires "identification mode" operations, then other recognition technologies such finger print, iris, retina etc, will probably need to be considered.  Advantages of using voice as a biometric includes:
Typical problems with voice recognition system include:
- Simple to use,
- feels natural to the user,
- provides eyes and hands-free operation,
- can easily be implemented to support remote recognition
- Implementation is usually inexpensive (often requiring software only).
As human voice is not rich in discriminative features such as fingerprints and iris patterns, there for voice recognition is a poor candidate for "identification mode" operation when there is a large database of enrolees.  Permanence of the voice recognition system can affect by change in voice due to aging and disease (e.g. stress, colds, and allergies). A system that updates the user's speaker model with each successful identification/verification might compensate for the changes.
- different type of environments (e.g. lab environment for enrolment, office environment for recognition )
- channel mismatch (e.g. different microphones for enrolment and verification)
- background noise
How Voice Recognition System works Like other biometrics systems, voice recognition systems also operate in two modes enrolment and authentication during the enrolment some sounds, words or phrases spoken by humans into a microphone are converted into electrical signals (from analog signal in to digital signals) by voice recognition system with the help of A/D converter  and then these signals are transformed into coding patterns (template) to which meaning has been assigned and store in to database which is referred to each time that person attempts to access secure data. When a user attempts to gain access to the secure data need to speak a phrase, the words are extracted and compared to previously stored voice models (template) and all other voice prints stored in the database. Each speech sound in the user's phrase is queried in an anti-speaker database. Since some characteristics of a person's voice are the same as another's, the system authenticates the user by comparing the user's common features with those in the anti-speaker database and eliminating those common elements from the sample to be authenticated. When all features matching others are removed, the system is left with only the unique features of the user's voice. These unique features, compared with the enrolled phrase, are the characteristics which determine successful authentication.
Main Features of Voice Recognition System
- Analysis of the unique characteristics of voice as a merger of physical and behavioural characteristics (physical dimensions of the voice box and the sounds adopted as speech is learned).
- Very little hardware is required (a microphone on a standard PC with software to analyze the unique characteristics).
- Ideally suited to telephone-based applications.
Biometrics Signature Signature verification is the process used to recognize an individual's hand-written signature. Biometric signature is a term used to refer to a signature that has been recorded/captured using a selection of input devices such as scanners, personal digital assistants, computer displays or other contact etc. This method allows real handwritten signatures to be incorporated into e-documents during electronic transactions. Not every technology captures signature information the same way. Signature capture is becoming fairly accepted as a replacement for pen and paper signing in bank card, PC and delivery-service applications. Biometric Signatures identification is also known as Dynamic Signature Verification (DSV) . This verification analyzes the way a user signs his/ her name. As signatures is used a means of transaction-related identity verification, among the peoples, most would see nothing unusual in extending this to cover biometrics. Signature verification devices are sensibly accurate in operation and clearly lend themselves to applications where a signature is an accepted identifier. In biometrics signature verification, important features of the finished signature such as speed, velocity, and pressure are use during the signature verification by verification devices. Verification devices use wired pens, pressure-sensitive tablets, or a combination of both. Devices using wired pens are less expensive and take up less room Currently, tablet-based systems that operate using off the shelf digitizers cost as little as Â£70-Â£100. Over 100 patents have been issued regarding signature verification to companies such as IBM, National Computer Register, and VISA. . although it's less expensive but are potentially less durable. So far, the financial community has been slow to accept automated signature verification methods for credit cards and check applications because signatures are still too easily forged. This keeps signature verification from being incorporated into high-level security applications. The major advantage of the Biometric signatures is if any process that requires a signature is a prime contender for signature identification. Individuals are less likely to object to their signature being confirmed as compared to other possible biometric technologies. Biometrics signature technology is also represents an ideal, bridge between the long-recognized convention of signing a document and the need for electronic documents to be uniquely recognized by individuals. This application provides individuals with security and control on documents originated, transacted and stored in the digital domain.
How Biometric Signatures works In Biometrics or Dynamic Signature identification systems the primary components of signature verification are specific feature of the process of the signature (behavioural component) such as stroke order, speed and pressure, and the difference of specific feature of the signature ( visual Images). There is an important distinction between simple signature comparisons and dynamic signature verification. Both can be computerized, but a simple comparison only takes into account what the signature looks like. Dynamic signature verification takes into account how the signature was made . Like other biometrics identification systems, in this system the devices convert the features of a signature in to a template and store it for future comparisons in their database. Signature identification devices can also analyze the static image of one's signature, which captures the complete image of one's signature and stores it for comparison. These devices account for changes in one's signature over time by recording the time, history of pressure, velocity, location, and acceleration of a pen each time a person uses the system. With biometric signatures, the authentication can be done in real-time or after the fact. In the event that a biometric signature is contested, the signature data can be extracted from the document and submitted to the verification devices for investigation and analysis to verify the authenticity of the signature.  The major problem faced with this technology is differentiating between the consistent parts of the signature and the behavioural parts of the signature that vary with each signing. An individual's signature is never entirely the same every time it is signed and can vary substantially over an individual's lifetime. Allowing for these variations in the system while providing the best protection against possible forgers is an apparent hurdle faced by this technology.
Main Features of Biometrics Signature
- The movement of the pen during the signing process rather than the static image of the signature.
- Many aspects of the signature in motion can be studied, such as pen pressure, the sound the pen makes against paper, or the angle of the pen that makes this a behavioural biometric.
- We learn to sign our name, and our signature is unique because of this learning process. The speed, velocity, and pressure of the signature remain relatively consistent.
Summary Each biometric product either under development or commercially available in the market can be categorized as falling into one of the biometric technology areas. The main biometric technologies are fingerprints, Iris, Retina, voice and digital signature. In finger print technology finger print is used as a biometrics sample while Iris and Retina are exam in Iris and Retina technology .In voice biometrics technology a pattern of voice is matched with previously store voice template in the database. The stroke order, speed, pressure and the difference of specific feature of the signature is verified in Biometrics signature system or Dynamic Signature identification systems.
Face Recognition system
General Background of Face Recognition system Face recognition systems identifies an individual by analyzing the unique shape, pattern and positioning of facial features. In other biometrics technologies, Face recognition technology is fairly young .it has been object of much interest in last few years. Facial recognition is also the most common method of human identification. It is non-invasive, usually passive, and fairly inexpensive, people generally do not have a problem accepting it as a biometric. It is probably for these reasons that face recognition has been one of the most active areas of biometric research. To improve and rebuts this technology, many companies pushing for the development of this biometric technology. Face recognition technology involves analysing certain facial characteristics, storing them in a database and using them to identify users accessing systems. There are various recognition methods that emphasize identification based on the areas of the face that don't change, including: upper sections of eye sockets, area surrounding cheek bones and sides of mouth. Making use of unique features or characteristics of the human face, often irrespective of facial hair or glasses, facial scan is deployed in fields as varied as physical access, surveillance, PC access, and cash point access. This system can also recognize faces within a crowd in the attempt to match them to stored images of known criminals. With the recent terrorist attacks there has been a tremendous push to implement this technology in a variety of public places such as airports, government buildings, border crossings, banks, public transport stations and other vulnerable areas. This technology is also use by law enforcement agencies in high crime areas like the streets of Tampa, London borough of Newham as well as other different cities. The use of this technology is also including identification systems for such things as security sensitive areas such as Nuclear power plants, cash points etc. This system is also use in many casinos to scan for cheaters and dishonest money counters. Face Recognition also provides the ability to reduce fraud and crime by identifying duplicate images in large databases, such as licensed drivers, benefit recipients, missing children and immigration. If the system is fully deployed it will contain up to 20 million images with the ability to retrieve images within seconds.  This biometrics base, Face recognition system can easily integrated with existing CCTV cameras with the help of software which can be used to monitor the criminals in area. The impatience to implement face recognition can be somewhat explained by its relatively cheap price. Recognition software can run as cheap as a few thousand pounds, and with the ability to utilize PC cameras the cost is significantly lower than other biometrics hardware.
Main Features of Face Recognition System
- Analysis of the unique shape, pattern, and positioning of facial features.
- Highly complex technology and largely software based.
- There are essentially two methods of capture: using video or thermal imaging. The latter is more expensive because of the high cost of infrared cameras.
- Primary advantage is that the biometric system is able to operate "hands-free," and a user's identity is confirmed by simply staring at the screen.
- Continuous monitoring of the user.
- Access to sensitive information can be disabled when the user moves out of the camera's field of vision.
- Verification is then performed when the user returns to work at the desktop.
How Face Recognition Systems works Face recognition technology involves analyzing certain facial characteristics, storing them in a database and using them to identify users accessing systems. There are various recognition methods that emphasise identification based on the areas of the face that don't change, including: upper sections of eye sockets, area surrounding cheek bones and sides of mouth. Like others Biometrics Technologies Face Recognition Technologies also operate in Enrolment and Authentication modes. During the enrolment mode several pictures are taken of one's face as a biometrics sample. This process will generally consist of a 20 -30 second . Ideally, the series of pictures will incorporate slightly different angles and facial expressions, to allow for more accurate searches. After enrolment, distinctive features are extracted, and converted in to mathematical codes called Template. This Template is then store to the data base. The template is much smaller than the image form which it is drawn. In verification mode(one to one operation), an individual approaches the checkpoint and presents an identity using a smart card, proximity card, PIN or other identity card have person image. The face recognition system take images, this captured images of the individual's face is converted in to a template (the mathematical process used by the computer to perform the comparison), after the template is generated, it is match with the store template for that person, if the score exceed from the set threshold then access granted. Threshold can be adjusted for different personnel, PC's, time of day, and other factors.
Duane M. Blackburn1 August 10, 2001 In identification mode (one-to-many operation) usually operate by comparing new unknown images with a set of known images, often referred to as the gallery or imge library. The unknown, or probe images, are usually obtained some time after those in the enrolled gallery. In identification setting the new image may belong to a person who is, or is not, already present in the gallery. Following three modules are the basic component of the face recognition system
- Face Detect
- Face recognition
Face Detection In order to make a perfect database for the recognition modules it is necessary to locate the exact position of the face in the image. Before recognition stage the face must be detected, extracted and normalized whether the input image is to be tested for identity or added into the database. The goal of face detection is to determine whether or not there are any faces in the image and, if present, return the image location and extent of each face . Face detection and normalization depends on each other. As each module requires the intermediate output of another module to continue with any further processing that's why they can not work individually. Two phases have been designed in face detection process i.e.
- Coarse detection phase
- Refined search phase
Course detection phase. In coarse detection phase a quick scan over the complete image is performed in order to analyse the colour content of the input. To locate the exact position of the face by applying refined search techniques in the second phase of face detection, which is performed after normalization of the luminance, the search space is reduced in this phase by identifying the skin regions in the image. Some basic steps are also involved in the face detection processes which are skin detection, Colour segmentation and ultimate erosion
Skin Detection Human skin colour has been used and proven to be an effective feature in many applications from face detection to hand tracking, although different people have different skin colour. Before entering into the face recognition system all the images or scenes from a video must pass through the first stage of the face detection module i.e. skin detection. In this stage a fast coarse search of the scene is performed in order to locate skinned regions in the image, so that the non-skinned regions can be removed. The purpose of this first stage is to perform a fast coarse search of the scene in order to locate probable skinned regions in the image, so that non-skinned backgrounds can be removed with the knowledge that the face will not be located in those regions of the image. A smaller image can then be extracted from the scene, such that successive searches for the precise location of the face can be performed on a reduced search space rather then the entire image. In this way the speed of processing and the accuracy of the location of the face are increased by removing the probability of the error and reducing possibility of FRR and FAR. Although skin colours vary between different people and different races but it has been verified that human skin tones form a special group of colours, unique from the colours of most other natural objects.
Colour Segmentation In Colour segmentation, first each pixel is classified as either skin or non-skin to perform skin region analysis. A down sampled version of the image is used in order to increase the speed of this module. For an input image of size 80 by 240, a down sampled rate of four is sufficient, such that the skin detection module only needs to operate on a 20 by 60 image . It has been observed that the accuracy of the search dramatically degrade if down sampling by a factor higher then four is applied . To classify the skin ness of each pixel a reference table is used, where each intensity is checked to see if it falls in the range of skin colour and a give it a binary value of one if it is and zero if not. In this operation(colour segmentation a bounded box is needed to determine the range and location of the values of ones, after the colour image has been mapped into a binary image of ones and zeros representing skin and non-skin regions. As the basic purpose of the colour segmentation is to reduce the search space of the following modules, there for it is essential to determine as tight a box as possible without cutting off the face. During the colour segmentation values that are nearly skin but non-skin or skin-like coloured regions that are not part of the face or the body are usually returned. These unwanted values which would be represented by a big connected region in the binary image are generally isolated pixels or group of pixels that are considerably smaller then the total face regions. The addition of these unwanted pixels would result in a box that is much larger than planned and overcome the reason of the segmentation. To reduce the effects of these unwanted pixels morphological refinements are applied to the binary output . Morphological techniques called erosion and dilations are used to eliminate these unwanted pixels which are generally smaller than the face region itself. Erosion followed by dilation is used in the system after the colour segmentation to clean up the binary mapping before extracting the skinned region.
Refined Face Search Luminance normalised skinned region of the image is the input of the second refined face detection stage. This stage is the final stage of the face detection module which involves a refined search for the location of the face. In refined search phase, depending on the status of the database two different algorithms are used. Normalised cross correlation is used to find out the centre location of the face when the database is empty and no faces have been processed by eigenfaces decomposition. In the training stage of the recognition module when a set of eigenfaces has been determined then the refined face searches can be applied by using a face projection search. The accuracy of the output from the refined search modules will decide how absolutely aligned the images are. Consequently, as the face recognition module requires perfect inputs that fulfil some restrictive criteria, the accuracy of these stages will determine how successful the recognition and the overall system will be. It is therefore very important and a high accuracy rate is of supreme concern. Two methods namely Normalised Cross Correlation (NCC) and Face Space Projection have been designed for this very important stage of the detection module depending on the status of the face database. Face space search involves projecting the sequence of windows into face space and measuring how "face-like" each window is while Normalised Cross Correlation deals with finding the best match between a template and a sequence of windows. Thus, the input data required and the choice of technique used is the major difference between the two searches and is dependent on what information is available. The Normalised Cross Correlation search requires a template; thus, a typical face or average face must be available in order for Normalised Cross Correlation to work. A set of eigen-faces is required by face space projection in order for each window to be projected into face space. The projection technique cannot be used until the set of dominant basis vectors has been calculated and available for use.
Normalised Cross Correlation Until a set of face images have been processed by the training stage of the recognition module the set of eigen-faces is not available. Template matching with Normalised Cross Correlation is used to perform the refined face search during the initial setup of the face database. Comparing two windows of the same size and determine its relationship or how closely linked each pixel from one window is from its corresponding pixel in another window is the vital theory behind Normalised Cross Correlation. Two windows under examination which has each of their corresponding pixels matching the closest from all other windows under testing are called maximized correlation value. Since it is the relative differences of intensity values within the picture which is significant during the matching therefore it is important to find a template that reflects the differing intensities of a face accurately. Thus, the best candidate for the template is the average face. It is a standard template representing the most essential features of faces and contains as little influenced additions as can be found of any face image. As it captures the relative differences in intensities between the features of a face it is therefore the most suitable choice for the template. An average face representative of the current ensemble of face images will not be available during the initial detection phase thus until the training stage, neither the eigen-faces nor an average face will be calculated.
Face Space Projection In face space projection technique each selected window is projected into face space. It is an accurate and alternative method to perform refined face search. A set of eigen-face templates can be determined by the training stage of the recognition module as soon as a database has been initialized by the Normalised Cross Correlation Given these templates, any subsequent image that passes through the second phase of the face detection module can hence utilize eigen-decomposition rather than normalized cross correlation. Like Normalised Cross Correlation (NCC) method, the input into this refined search stage which is the normalized skinned region of the original image is treated as the test window. During the search, sections of this test window the size of an eigen-face will then be continually extracted for projection into face space. The primary step is the loading of the average face and the eigen-faces that were saved from the training stage of the recognition module. Notice that the average face used here is the actual average calculated from the set of input faces that were added into the system, unlike the average face used in Normalised Cross Correlation (NCC), which is foreign to the current database. The projection is accomplished by subtracting the selected section of the test window by the average face when the average face and eigen-faces are available. Since all eigen-faces are calculated from the covariance matrix, which originate from zero mean images therefore it is essential pre-processing step to transform the data into a zero mean image In order to produce a set of weights the zero mean test region is then projected into face space by multiplying it with the loaded set of eigen-faces. Using the weights, by comparing the energy of the window to the energy of the transformed window it is likely to determine how "Face-like" the region is. Large weight values will be recorded and the amount of projections onto face space would be maximised if the region was face-like" since every of the basis vectors capture the main variances in a face. Since the eigenvectors do not clearly represent images that do not reflect a face structure a "non-face-like" region will therefore produce a smaller projection values. The maximum value that the weights can have is theoretically the total energy of the original window itself prior to transformation, that is, every bit of energy is conserved and transformed into face space.  As mathematically energy is the sum of the squares of the intensity values. Theoretically, the difference between the sum of squares of the transformed window i.e. the weights and the norm of the original window should be zero if the region was exactly centred on the face. Every particular region from the test window will therefore have an related distance from face space recorded, such that the least distance out of all the regions tested will symbolize the nearby matched, the most "face-like" area, in the test window. While operating in dynamic mode Face detection is completed in real-time, repeatedly processing images. Therefore not only the speed but also the accuracy of this refined search module is important. The quicker the face space search can be, the more extraordinary the performance of the system can be contrary to the Normalised Cross Correlation (NCC) module, which is applied at initialization. To attain such effects a multi-resolution search is therefore employed here. The image is first down sampled where a rapid coarse search can be applied prior to inputting the normalized skinned region into the refined face space search. Now to locate the exact centre of the face, neighbouring pixels around the first estimate will be tested and which will improve on the accuracy of the search. With the amount of speedup limited by the accuracy of the down sampled window, the unrefined to fine search technique used which has provide significant speed improvements.
Face Normalization As face detection and face normalisation are interdependent therefore the success of the overall system depends to a great extent upon how these two modules work mutually. For the success of face detection the input faces must be normalized so that the lightning and orientation of the images become similar to the stored templates of the system during the enrolment mode that is all input images are captured under the exact same conditions and environment. This is the main problem for many research groups in this area that's why comprehensive solution to the problem of face recognition has not been presented. Face normalization module can be divided into following parts
- Lighting normalization
- Scaling normalization
- Rotation normalization
- Background subtraction
Lighting Normalisation In lightning normalization the luminance of the images are adjusted in such a way that it can be regarded as taken under the same lightning conditions. Careful concentration needs to be paid in conserving the equal amount of luminance on all images so that no particular bias is placed upon lighting differences because the refined face search modules greatly depends on the intensity values on each pixel. The recognition should not be based upon the background lightning conditions or the time of day that the image was captured but upon the variances and the relationship between faces. The total amount of lightning can be normalized by observing the energy embedded in the images. The relative ratios between the pixels within the image itself must not be changed for unbiased comparisons between images. Thus for the resultant total energy to be a standardized value all intensity values within the image should be multiplied by a constant ratio. The ratio of energies determines this ratio. There are two places where the lighting normalization module is applied i.e. before the refined detection module and prior to entering the face images into the recognition module. To improve the accuracy of locating the face the first normalization is applied in such a way that all selected regions have the same total energy as the average face In order to avoid extreme biasing towards a particular lighting condition in the image when the distance functions are calculated.  since it would naturally record a higher correlation value regardless of the template when light is incident from one end of the image to the other, and biasing is placed upon the brighter side and this was the problem with normalised cross correlation. This problem is relieved by the addition of lighting normalisation stage. For all images inputting into the recognition module to have the same total energy, so that no bias is placed upon the particular face during recognition second lightning normalization stage is applied. A conservation of energy would act as if all images of the heads were taken under the same lighting conditions because all images at this stage will be the same size The total energy of an image is the sum of the squares of the intensity values. Which makes the ratio between intensities a square rooted relationship between two images. Energy = (Intensity Values)2  Consider a face recognition system where all images are stored as intensity values. Thus, in the luminance normalisation module the before comparing it with the energy of the average window the energy of the selected window will first be calculated after that the ratio of the energy is computed then the intensity of each pixel in the window will be multiplied with the square root of that ratio. Thus the total energy of the image is normalized in the resultant image.
Scaling Normalisation For the proper operation of the two refined search stages a standardized scale is necessary. It is a condition that the face in the input image must matches the size of the faces stored in the database for an accurate face detection; and for recognition unless the faces are matched in the scale it is not possible to compare it. If the faces are being photographed at different distances from the camera then the scaling problems will arrive and therefore the faces need to be normalized to a standard size required by the database if the face is smaller due to larger distance from camera or bigger because the face is being closed to the camera. Scaling problems have affected in it-self many proposed methods of solving scaling problem, such eyes detection. Thus, another method is proposed for this system which does not depend on correct scaling in the first place. This method fully utilized the skinned region information obtained from the colour segmentation stage. The size of the extracted box is compared to the dimensions of the template depending on the binary mapping of the face region in the image. The template will be the average face obtained and face space projection and face recognition will use the eigenfaces as the template in case of normalized cross correlation. Two scaling factors, one being a ratio of width and the other, a ratio of height are then determined. The larger ratio between the two ratios is selected when scaling down and the smaller ratio is selected when scaling up. In order not to morph the face to a dissimilar set of dimensions by maintaining the relative variances between the pixels constant the process of using same ratio for both dimensions is chosen. Normalisation can be achieved either by adjusting the dimensions of the extracted box, or by extracting a new box from a scaled version of the image once the ratio has been determined, such that the face in the image is now in the same scale as the templates. To locate the centre of the face refined search techniques such as face space projection can then be applied.
Rotation Normalisation The faces in the scene will normally be subjected to different rotational orientation in case of video sequence. A face recognition system would need to recognize faces without placing limitations upon the acceptable rotational orientations of the faces In order to match the recognition capability of human beings, which is currently under vital research. The major problems come across by face normalization modules due to three dimensional rotary motions of heads are forward tilting, planar rotations tilting of and turning rotations as shown in figure. In this system because of the complexity only planar rotation has been investigated. A process concerning the use of the skin detection binary image to report for rotation has been proposed for this system. The main purpose is to calculate the angle of rotation of the face block in the binary image. To achieve this ellipse is wrapped around the faced region so that the second-order moment of the ellipse is the same as the second-order moment of the binary mapping. Thus the face's angle of tilt will be the angle formed between the major axis of the calculated ellipse and the horizontal x-axis. By rotating the image by the negative of that angle Rotation normalisation can be achieved 
Background Subtraction The extracted face input into the recognition stage is a rectangular box around the face under the current design of the face detection and normalization modules. Because a face is elliptical in shape, no matter how firm a surrounded region is determined, with a rectangular box, there are bound to be sections of the background incorporated inside the extracted image. Therefore the main purpose of the background subtraction is to improve the face inputs by removing the background entirely so that only the face is present in the extracted image, by using the colour segmentation results of the face detection stage, using the binary image as a mask over the face region this can easily be accomplished. For the inclusion of a background subtraction stage into the design of the recognition system considerations were made, and experiments and tests were carried out to assess the usefulness of this addition. The results showed that the recognition performance decreased considerably with the addition of a background subtraction stage due to the extreme quantity of energy being placed into mapping the boundaries and edge effects of the cutting. Hence it is measuring the shape of the background segmentation rather than recognizing the variances of the face. Consequently this method is practically unfeasible and has not been integrated into the system although this method theoretically enhances the recognition.
Systematic Block Diagram of the Real Time Face Recogniton System
Face Recognition Achieving best results form faces recognition system, it's important to Provide a perfectly aligned standardized database is available, the face recognition module is the most reliable stage in the system. As in face recognition lies in the normalization and pre processing of the face images so that they are suitable as input into the recognition module. Many researches have been made in this filed, these literatures addressed the issues of recognizing faces based on a pre-processed face database. Given a perfect set of faces such that the scale, rotation, background and luminance is controlled, the recognition module can work with the best possible ideal inputs, since it is crucial that the performance of this foundation module be as optimized as possible. Its ability to recognize an ideal database will determine the best possible performance attainable by the overall complete system. Any subsequent development and implementation of the face detection and normalization module will therefore be aimed at providing this ideal set of database. Following are the different methods which can be used in face recognition.
There are four primary methods used to recognize a user face.
- Feature analysis
- Neural network
- Automatic face processing
- Methods of Face Recognition
- Feature analysis
- Neural network
- Automatic face processing
Eigenface Eigenface idea 1st proposed by Sirovich and Kirby  which is later on refined and extended by Turk and Pentland  by adding pre-processing and expanding database. This method is based on the principal component analysis. The idea is to reduce each face image to a vector, and then uses Principal Component Analysis (PCA) to find the space of faces. This space is spanned by just a few vectors (Eigenvector), which means each face can be defined by just a set of coefficients weighting these vectors. This approach can be applied both on detection and recognition. The small set of characteristic feature images called eigenfaces, which may be consider of as the principal component of the original images. This is important because it allows us to explain an observation with fewer variables. This not only scales each variable according to its relative importance in explaining the observation but also decreases the computational complexity of face recognition Variation between facial images in an orthogonal basis set of vectors (eigenvector) is capture in this method. The eigenvectors are thus the image vectors, which map the most significant variation between faces. The accuracy rate for this face recognition system is about 70%. If the training data contained not only non-smiling faces, the results would be higher. For some false cases, even humans are not able to tell on a 25 by 25 pixel image. 
Algorithms of Eigenfaces Exclusion of eigenfaces is necessary because of the shortage of computational resources. It is possible not only to extract the face from eigenfaces given a set of weights, but also to go in the reverse mode. This reverse mode would be to extract the weights from eigenfaces and the face to be recognized. These weights give us the amount by which the face in question is different from typical faces represented by the eigenfaces. These weights can be used to determine two important things.
- Similar faces (images)
Determine Determine if the image in question is a face or not. In the case the weights of the image is different to a great extant from the weights of face images (i.e. images, about which we are sure that they are faces), the image probably is not a face.
Similar faces (images) Similar faces have similar characteristics (eigenfaces) to similar degrees (weights). If one extracts weights from all the images available, the images could be grouped to clusters. That is, all images possessing similar weights are probably similar faces.
Eigenvectors and Eigenvalues In Matrix an eigenvector is a vector of a matrix, if it is multiplied with the matrix, the result is always an integer multiple of that vector. This integer value is the corresponding eigenvalue of the eigenvector. This relationship can be described by the equation Ã— u = M Ã— u, where u is an eigenvector of the matrix M and is the corresponding eigenvalue. 
Feature analyses Human Being recognise people mainly base on features such as eyes, nose and mouth, and there spatial arrangement. Feature analysis is also called Local Feature Analysis (LFA); it is the most widely used technique because of its capability to accommodate for facial changes and characteristic.
How Local Feature Analyses works Local Feature Analysis (LFA) is based on local features that make up part of the face that distinguish between persons. These features combine the texture information of a face image with information of the underlying anatomical bone structure. Identity is determined by combining information of these features in a statistically meaningful way .LFA is part of Visionics' FaceIT, which uses an algorithm to create a face print (84 bytes in size) for comparison. 
Neural network Neural Network analyses features from both images, the enrolment and verification image and determines if there is a match using an algorithm.
How Neural Network-Based System works Neural Network-Based Face Detection operates in two modes; in first stage a set of neural based filters is applied to an image, and in second stage an arbitrator is used to combine the output (result). To locate a face the filters scan each location in the image at several scales. The arbitrator then merges detections from individual filters and eliminates overlapping detections. In this system a filter receives an input as 20x20 pixel region of the image, and generates an output ranging from 1 to -1, indicating the presence or absence of a face, respectively . The filter is applied at every location to detect faces anywhere in the input. The input image is repeatedly reduced in size and the filter is applied at each size in order to detect faces larger than the window size. This filter must have some invariance to position and scale. The quantity of invariance determines the number of scales and positions at which it must be applied. Here the filter is applied at every pixel position in the image, and the image is scaled down by a factor of 1.2 for each step in the pyramid.  As shown in Fig. 1.To decide whether the window contain a face or not a pre-processing step is applied to a window of the image and the window is then passed through a neural network. The pre-processing (as shown in the fig 2) first balances the intensity values in across the window. A linear function is then applied across the window to the intensity values in an oval region inside the window. To compensate for a variety of lighting conditions, the linear function will estimate the overall brightness of each part of the window, and subtract it from the window. To increase the range of intensities in the window histogram equalization is performed this non-linearly maps the intensity values. The histogram is calculated for pixels inside an oval region in the window which improve contrast in some cases as will as compensates for differences in camera input gains. The pre-processed window is then passed through a neural network. The neural network is connected to its input layer units. There are three types of hidden units; the receptive fields of hidden units are shown in Fig. 1. Each of these types was selected to allow the hidden units to identify local features that might be important for face detection. The features such as mouths or pairs of eyes are detected by the horizontal strips of the hidden units while the features such as individual eye, the nose, or the corners of the mouth are detected by the square receptive fields of the hidden units. The outputs are then combined and the threshold is then applied.
Automatic Face Processing Automatic Facial Processing (AFP) is the simplest method of facial recognition. In this method of facial recognition facial detection methods are used to locate and measure facial features such as the eyes, nose, cheeks, mouth etc. To produce distance ratio, the distance between other features and the size of these features is then calculated. The facial code is then created by using the resulting string of ratios which is used to store and search database. The facial database is divided into classes to accelerate the search process. For example all people with 6 cm mouths may be stored in the same class thereby reducing the time required to match a person with a 6cm mouth and a 4cm nose.
Summary Facial recognition technology, part of the biometric collection of security systems, is used in security applications such as gaining system access and entry after verification. Its also enables the scanning of faces from live video, with standard security cameras, and matches each face that appears in the field of view against a watch list database. Facial recognition technology works by establishing a diagnostic dimension of the landmarks of a face, no matter which profile, and how they fit together. Change in appearance such as a moustache, glasses, and even standard plastic surgery does not alter the essential landmarks of a face, the original face can still be found, and the original identity ascertained. Face recognition in general and the recognition of moving people in natural scenes in particular, require a set of visual tasks to be performed robustly. These include Face detection: The detection and tracking of face-like image patches in a dynamic scene, Normalisation: The segmentation, alignment and normalisation of the face images, and Recognition: The representation and modelling of face images as identities, and the association of novel face images with known models. Several methods that can be used for face recognition. Some of the most common are using PCA or Eigenfaces, Local Feature Analysis, Neural Network and Automatic Face processing.These tasks (detection, Normalisation, Recognition) seem to be sequential and have traditionally often been treated as such. However, it is both computationally and psychophysically more appropriate to consider them as a set of co-operative visual modules with closed-loop feedbacks.
Biometric Access control on different campuses of University of East London, Using facial recognition to fight Attendance fraud Attendance fraud in colleges is a serious problem and has a great impact on the overall academic performance of students of a University and which also affect the rate of ranking of a university. In order to solve this problem in an efficient way a biometric based facial recognition system can be introduced in Universities which will comprehensively control the student's presence and study hours on the basis of automatic facial recognition. Almost every university has to cope with the problem of "black sheep" among the students. Some students secretly stay away from classes while still being registered as attending by asking a colleague to register them. Other people could come in and out of the university as will and could pose a security problem. In order to expose such methods of deceit, an entirely new procedure can be introduced in universities, such as the University of East London. Students could have to verify themselves not only by means of their student ID but also by their facial features.
From a Traditional time registration system to biometric access control A system can determines by means of facial recognition that whether the card swiped by a student before entering into the class really belongs to that student. Before starting a class, every student could present "his/her face" to the recognition camera. Consequently, deceit of the system by a "change of identity" is completely ruled out.
Biometrics in practice - use of facial recognition in a university The employment of facial recognition in a university could be guided by the following questions: "Can the system be combined with time registration devices?", "How can we increase the users' acceptance?", "How much time is needed for registration?", "Is the system easy to operate?", "Does it require high maintenance?", "Where can the system be installed?" etc. The installation of a facial recognition system could be closely coordinated with these questions and be carried out in adjustment to the surrounding conditions set by the university. Via an interface in form of a log file, the system could be easily integrated into the attendance system. In detail, the integrated process of attendance system and facial recognition could be carried out as follows: When the identity card is read into the system, this simultaneously triggers the biometric verification procedure. The integrated facial recognition camera records the face of the person that is verifying himself or herself, and within fractions of a second, the live image is compared to the reference data stored in the system (personal data and face). Only if the number of the compared facial features which match is sufficient to guarantee the students/ employees identity, the facial recognition system automatically registers the start of the attendance hours. Any failed attempts at registration are recorded by the system as well, which means that the university can precisely detect any attempted deceit. From the very beginning, the knowledge of this log file within the facial recognition system may prevent any manipulation from the part of the students and any other person. Facial recognition technology, that bases its comparison of the live image and the reference image on a total of up to about1, 500 facial features, is very robust and offers a high capability of abstraction, thus safeguarding a constant recognition performance even in case of changes to the outward appearance such as a changes in hair style or presence or absence of bear. Such system's performance is not even reduced when scarves or protective clothing are worn, as it unavoidably happens during everyday verification. However, not only its reliable recognition performance, but also the acceptance by the students may be a precondition for the installation of a facial recognition system. If the users do not accept a system and thus are not willing to cooperate, this directly affects the recognition results. The safest recognition system is useless if the relevant persons refuse the verification process. Since the facial recognition process is carried out contact less, it does not restrict the student's freedom to move, and the quick facial comparison will not require more time than the already existing time registration process familiar to students. In addition, the operation of the system can be easily learned by everybody. Thanks to the high speed of registration, it is possible to carry out the regular control registrations during the day demanded by the university. These control registrations by means of a facial recognition system could enable a continuous survey of the presence of all students on the campus without undesirable loss of time. This can certainly help if, for example, evacuation is required due to a fire. The fact that the biometric data are registered contact less has another significant advantage: the maintenance required for the system is minimised because the contact areas cannot become soiled, unlike e.g. finger print sensors. This system can also be used for security purpose at university campuses e.g. at UEL Barking and Dockland LRC entrance and student residents and other secure places i-e Network control rooms and different sensitive labs, if this system is installed it will not only stop impostor students who use other students IDs but will also replace the security guards to whom the university is paying a huge amount. This (facial recognition) system can also be used with the university payroll system for university employee. This system can also be used to secure individual account of every student from unauthorized access by installing it on every computer in laboratory.
Problem with Existing System
- Lack of control over students who leave their classes after the registration in the morning or have themselves registered by colleagues,
- Traditional attendance system offered only insufficient control as it cannot verify whether an student card really belongs to an employee, which means that a "change of identity" is possible
- A solution for attendance registration and unambiguous identity verification into one overall system needs to be searched.
- Automatic identity verification based on automated facial recognition, Integration of the product solution of a facial recognition system into a suitable time registration system via an ODBC (Open Database Connectivity),
- By presenting the student ID entering to university building , the quick and contact free facial recognition process to verify the student's identity is triggered,
- Only if the compared facial features match, will the facial recognition system automatically register in the verified student,
- Regular control registrations carried out by the facial recognition system checks the presence at the university, each day's last control registration being automatically recorded as the end of the day.
- A facial recognition system can run extremely stable and could provide a reliable control of the students presence and studying hours
- The system is easy to operate and should enjoys a high acceptance by the students who verify themselves via the automated facial recognition several times a day
- Facial recognition system scan be easily integrated into the given surrounding conditions and optimises the everyday routines on the university.
Characteristics of the proposed system As there are different types of facial recognition system available in market, it is important before selecting any system for organization; the system must available for single and multiple stations and can be fully integrated in existing access control and time attendance systems.
User friendly The system must be friendly user, understandable. There must not be any treat to the users. During the recognition the system must recognised user in the same way that a human doorman would recognise some one. The system must operate without physical contact. Change in facial feature must not effect the performance of the system e-g they must recognises user with or without a beard, and cannot be tripped up by a new pair of glasses or a change in expression either.
Secure The system must preclude any attempts at trickery with photos and masks. The system should record all events so that these events can be checked at anytime be the system administrator.
Economical In order for the system to be economical it should be rapidly installed, even in existing access control or time clocking systems. Enrolment of the new student should be simple. By simply taking one or two pictures and that could be taken at any station or at a special photo location.
Standard Interface support The system must support the standard protocols interface, There must be support for the ODBC database interface standard which easily allowed the acceptance of the master data from external databases.
Scalability The system should accommodate huge amounts of data and can use the latest dual processing. Such a system accommodates large amount of memory to provide you with far greater efficiency in managing high-bandwidth input/output .
Live Video Capture The system must provide a complete live video face recognition solution, Separate Authentication and Identification modes as well as motion detection and scene verification modes.
System Specifications The following specification of facial recognition system was complies after extensively study of the ZN Technology AG, IMAGIS Technology INC, CCESOFT Ltd and Viisage Technologies.
Conclusion Imagine a world where trust can be taken for granted. Security is seamless. Imagine a world where trust can be taken for granted. Security is seamless. Preventive measures are superfluous. Regrettably, we don't live in that world today. Our news is filled with reports of security breaches, terrorist plots, heinous crimes, computer hackers, online fraud, and identity theft. Global security is more important than ever. The concept of individuality of personal traits has a long history, and identification of a person based on his physical characteristics is not new. Human or even animals recognise each other based on their physical characteristics. Just over decay ago, the idea of biometrics technology can only be found in Hollywood films or research paper; however, this very same technology is now increasingly being used on access control products. They provide greater security, conveniences over conventional control mechanisms, and the public is becoming award of its potentials. Given the requirement for determining people's identity, the obvious question is what technology is best suited to supply this information? There are many different identification technologies available, many of which have been in commercial use for years. The most common person verification and identification methods today are Password/PIN known as Personal Identification Number, systems. The problem with that or other similar techniques is that they are not unique, and is possible for somebody to forget loose or even have it stolen for somebody else. In order to overcome these problems there has developed considerable interest in "biometrics" identification systems, which use pattern recognition techniques to identify people using their characteristics. Some of those methods are fingerprints, retina, iris, voice and face recognition Face recognition technology has come a long way in the last few years. Today, machines are able to automatically verify identity information for secure transactions, for surveillance and security tasks, and for access control to buildings. These applications usually work in controlled environments and recognition algorithms that can take advantage of the environmental constraints to obtain high recognition accuracy. However, next generation face recognition systems are going to have widespread application in smart environments, where computers and machines are more like helpful assistants. A major factor of that evolution is the use of neural networks in face recognition, neural networks more efficient, popular and helpful to other applications. Applications for face recognition technology can be found in many government and industry areas because this technology is the most convenient and natural way of recognizing people, and at the same time it is one of the most reliable biometric methods. Specifically border control, there are various opportunities where face recognition can help to make a country border more secure. At their current level of development, facial recognition systems show promise but are not yet advanced enough to be considered mature technologies, its still need to be improve.
Future Works Although face recognition systems can work fine under controlled conditions like frontal mug shot images and regular lightning but it fails under the enormously changing conditions in which humans can and must recognize other people. Next generation face recognition system will have to identify people in real time and in much less controlled situations. I believe that face recognition systems that are robust in usual environments (in the presence of noise and light changes) cannot depends on a single modality; therefore, combination with other modalities is necessary. Technology used in elegant environments has to be inconspicuous and permit users to act without restraint. Considering all the needs, systems that use face and voice recognition seem to have the most prospective for wide-ranging application. Cameras and microphones today have been incorporated with wearable systems because of its small size and lightweight. The significant advantage of audio and video based recognition systems are that it uses the same modalities that are used by humans for recognition. The objective of elegant environments is to create such an environment where computers and machines are more like useful assistants, rather than lifeless objects. Face recognition technology could play a major part in achieving this goal. But to achieve the goal of wide-ranging application in elegant environments, next-generation face recognition systems will have to easily fit within the outline of standard human interactions and match to human intuitions about when recognition is expected. Finally, researchers are beginning to demonstrate that inconspicuous audio and video-based personal identification systems can achieve high recognition rates without requiring the user to be in a highly controlled environment. Even though significant research remains to be done, these targets now appear to be within reach. Although there is some achievement in 3D based face recognition system design, but it still need to improve for real time recognition in real world. This will automatically extract feature. Future work also includes the choice of components, pre-processing techniques and training mechanisms
ISSUES Great technology requires great responsibility. Nowhere is this more applicable than the field of biometrics. On their own, biometrics are neutral technologies, vital to establishing trusted identity, preventing ID fraud and protecting critical assets and sensitive information. However, biometrics is also a form of personal information, and should be treated as the most appropriate and ethical ways for society to benefit from its technology while helping to minimize any opportunities for abuse. There must be formulated responsible use guidelines, which secured user acceptance by those adopting Biometrics technology, been vigilant in ensuring compliance and, where possible, its must have built technical measures to maintain control over the installations. One of the most crucial factors in the success of a biometric system is user acceptance of the device. The biometric devices must work correctly. When it functions properly, it does two things: it keeps unauthorized people out and let's authorized people in. The system must not discomfort the user. If people are afraid to use the device, they probably will not use it properly, which may result in users not being granted access. The contrast and the frustration of dealing with a high number of false rejects will have authorized users and management alike looking for a way to replace the biometric system with something else if these factors are not considered up front. One of the most crucial issues is the implementation of face recognition system in counties like Afghanistan where most of the women cover their faces completely and do not show their faces to any one except their close family members ( husband, brother, father ). As Biometric measurements deals with complex patterns that reveal certain inconsistency introduced by either the measurement or tested attributes itself. For example one's handwritten signature varies every time he/she write it, one's voice may sound different, or one's fingerprint digital representation might be changed by a cut. Thus biometric identification schemes can be affected by selecting high threshold. This susceptibility must be controlled by careful risk assessment and system validation. Attacks on the security of biometric systems are most likely to concentrate on different vulnerabilities in the security chain. The biggest danger with biometrics is that once an individual's biometric data or parameters are stolen, they are compromised for life. That is very different from the example of the stolen key. A biometric cannot be updated. In this case, disclosure of template data stored at any biometric authentication server could compromise use of that biometric technique for the affected users, forever. To overcome the limitation of the single identifier biometrics based identification system it is recommended that in highly sensitive areas Multimodal biometrics base identification or single identifier biometrics base identification system with existing authentication methods should be used.
A Suggestion for UK Immigration Authority There is an increasing demand from immigration and customs authorities to automate the border control process, at the same time as ensuring secure checking of travel documents and verifying the identity of the passport holder by a biometric system. There are many ways where illegal immigrants enter the UK. An example of this is using a passport, which does not belong to the individual, without changing the photo. Many ethnic groups share common facial features and it can sometimes be difficult for an immigration officer to differentiate between the photo on the passport and the impostor. This can lead to the immigration allowing illegal immigrants into the country. To overcome this it is strongly recommend that the UK government employ the use the facial recognition system to prevent illegal immigrants entering to UK. This can be achieved through the use of the facial recognition system being used in a slightly different way from the enrolment system. A picture of the person could be captured through a camera and could be analysed through a computer system to match passport photo. The system will then verify whether the passport belongs to the individual being assessed. It will not only reduce work loads on immigration officers but also reduce the possibility illegal immigrants entering the country.
References and Further Readings
- ASHBOURN, J. Biometrics: advanced identity identification: the complete guide. Springer-Verlag, London, 2000.
- Biometrics Consortium http://www.biometrics.org
- International Biometrics Group: http://www.biometricgroup.com
- M.Bellare, A.Desai, E.Jokipii and P.Rogaway, "A concrete Security Treatment of Symmetric Encryption" IEEE computer Society Press 1997
- "Automated biometrics-based personal identification" Proc. Natl. Acad. Sci. USA Vol. 96, pp. 11065-11066, September 1999
- International Biometrics Group "research consulting Integration, Biometrics FAQ" Version 1:0, 2001
- "National Biometrics test centre collection works", by James L. Wayman, Director, version 1.3, August 2000.
- A.K. Jain, L.Hong, Y. Kulkarni, "A Multimodal Biometric System using Fingerprints, Face and Speech", 2nd Int'l Conference, March 1999.
- "A piratical guide to biometrics security technology" by Simon Liu and Mark Silverman January 2002.
- "Industry insight and analysis for Biometrics marketplace" Volume 02, issue March 2003
- "The fight for privacy has just begun" business week, Jan 10 2002.
- "modernism government" UK biometrics working group issue 1.0, 23rd November 2001
- "Banks eye biometrics for ATMs, increasing security concerns and fraud renew interest password methods" by Lucas Mearian, computer world, Monday ,January 14, 2002
- "Enhancing security and privacy in biometrics-based authentication systems" by N.K Ratha, J.H.Connell, R.M.Bolle
- "Let your fingers do the logging in" by Willis, David (Network Computing, 06/01/98
- "The Guardian", (newspaper) Friday July 25 2003
- "Metro" (news paper), Thursday, July 10 ,2003
- "An experimental comparison of secret based user authentication technologies" S.M Funell, Network Research Group, Plymouth University.
- A.K. Jain, R. Bolle and S. Pankanti (Eds.), BIOMETRICS: Personal Identification in Networked society, Kluwer Academic Publishers, 1999.
- S. Prabhakar, S. Pankanti, and A. K. Jain, "Biometric Recognition: Security & Privacy Concerns", IEEE Security & Privacy Magazine, Vol. 1, No. 2, March-April 2003.
- Y. Wang, T. Tan and A. K. Jain, "Combining Face and Iris Biometrics for Identity Verification", Proc. of 4th Int'l Conf. on Audio- and Video-Based Biometric Person Authentication (AVBPA), pp. 805-813, Guildford, UK, June 9-11, 2003.
- L. Hong and A. Jain, "Automatic Personal Identification by Integrating Faces and Fingerprints", Proc. IEEE Workshop on Automatic Identification Advanced Technologies, pp. 15-18, Stony Brook, NY, November, 1997.
- D. Maio, D. Maltoni, R. Cappelli, J. L. Wayman, and A. K. Jain, "FVC2000: Fingerprint Verification Competition", IEEE Transactions on PAMI, Vol. 24, 2002.
- "Biometric Personal Identification System Based on Iris Analysis." U.S. Patent No. 5,291,560 issued March 1, 1994 (J. Daugman).
- "How Iris Recognition Works" John Daugman, PhD, OBE
- R.Bright, smartcards: Principles Practice, Application.New Yark Ellis Horwood, Ltd 1998
- "Retina-Scan" by Slashdot, 1999
- F.H. Adler, Physiology of the Eye: Clinical Application, Fourth ed. London 1965.
- "Speaker Recognition: A Tutorial," by J. Campbell, Proceedings of the IEEE, Vol. 85, No. 9, September 1997.
- "Automatic Identification and Data Capture", http://www.tech.purdue.edu/it/resources/aidc/BioWebPages
- "Automated Biometrics Technologies and Systems," by D. Zhang, Kluwer Academic Publishers, Boston, 2000.
- "Finger Pressure", Bioscrypt Inc, Security Systems industry,, forecasting Canada Canadian Business, 08/06/2001, Vol 74
- L. Sirovich and M. Kirby, "Low dimensional procedure for the characterization of human face image", Journal of Optical Society of America, vol.4,
- "Enhancing assistive technologies: through the theoretical adaptation of biometric technologies to people of variable abilities" bye William J. Lawson, Ph.
- "Natural Image Correction by Iterative Projections to Eigenspace Constructed in Normalized Image Space" By Takeshi Shakunaga, Fumihiko Sakaue
- "Design of Real Time Face Recognition System" by S.K.Singh, D. S. Chauhan, Mayank Vatsa, Richa Singh, dated 03/01/2003.
- "A Survey of Face Recognition" , By Thomas Fromherz, Peter Stucki, Martin Bichsel
- Rowley, Baluja, and Kanade: "Neural Network-Based Face Detection" (PAMI, January 1998)
- "Color Image Segmentation: A State-of-the-Art Survey", By L. Luccheseyz and S.K. Mitray
- "Visual Speech for Speaker Recognition and Robust Face Detection", By A. Pentland, B. Moghaddam, and T. Starher
- "Face Detection in Color Images", By Hsu, Abdel-Mottaleb, Jain (2002)
- Automatic Extraction of Lip Feature Points Roland G", J Bruce Millar, Alexander Zelinsky, and Jordi Robert-Ribes
- Probabilistic Tracking in a Metric Space", Microsoft Research Ltd
- Background and Foreground Modeling Using Nonparametric Kernel Density Estimation for Visual Surveillance, By Ahmed Elgammal, Ramani Duraiswami,
- "Is All Face Processing Holistic", By Garrison W. Cottrell, Mathew N. Dailey
- "Learning the statistics of people in images and vedio", By Hedvig Sidenbladh
- R.L. Klatzky and F.H. Forrest, "Recognising Familiar and Unfamiliar Faces," Memory and Cognition, Vol. 12, 1984.
- A.C. Schreiber, S. Rousset, and G. Tiberghien, "Facenet: A Connectionist Model of Face Identification in Context," European Journal of Cognitive Psychology, Vol. 3, 1991.
- M. Turk and A. Pentland, "Eigenfaces for Recognition", Journal of Cognitive Neuroscience, vol. 3, no. 1, pp. 71-- 86, 1991.
- L. Sirovich and M. Kirby, "Low-dimensional procedure for the characterization of human face image," J. Opt. Cos. Amer., vol 4,
- J. Daugman, "Wavelet demodulation codes, statistical independence, and pattern recognition," Institute of Mathematics and its Applications,
- "N-Feature Neural Network Human Face Recognition", Javad Haddadnia, KarimFaez, Majid Ahmadi, Amirkabir University of Technology,
- "Detecting Faces in Images: A Survey", IEEE Transactions on Pattern analysis and machine intelligence, vol. 24, no. 1, January 2002
- Viisage's Technology, (Viisage's Face Explorer) http://www.viisage.com/facexplore.htm
- P. K. Janbandhu, M. Y. Siyal, "Novel Biometric Digital Signatures for Internet based Applications" Information Management and Computer Security (journal), Emerald, vol. 9
- "Biometrics and Digital signatures in Electronic commerce", by R. R. Jueneman, R. J. Robertson, Jr.
- A. Jain, R. Bolle, S. Pankanti, editors, "BIOMETRICS Personal Identification in Networked Society," Kluwer Academic Press, Boston, 1999.
- M. Turk and A. Pentland, "Eigenfaces for Recognition," Journal of Cognitive Neuroscience, Vol. 3, Mar. 1991
- A.W. Young and V. Bruce, "Perceptual Categories and the Computation of 'Grandmother'," European Journal of Cognitive Psychology, Vol. 3, 1991
- J.W. Shepherd et al., "The Effects of Distinctiveness, Presentation Time and Delay on Face Recognition," European Journal of Cognitive Psychology, Vol. 3.
- R.L. Klatzky and F.H. Forrest, "Recognising Familiar and Unfamiliar Faces," Memory and Cognition, Vol. 12, 1984.
- A.C. Schreiber, S. Rousset, and G. Tiberghien, "Facenet: A Connectionist Model of Face Identification in Context," European Journal of Cognitive Psychology, Vol. 3, 1991.
- S. Lawrence et al., "Face Recognition: A Convolutional Neural Network Approach," IEEE Transactions on Neural Networks, Vol. 8.
- T. Kohonen, Self-Organization and Associative Memory, Springer-Verlag, Berlin, 1989.
- G. Cottrell and M. Fleming, "Face Recognition using Unsupervised Feature Extraction," Proc. Int'l Neural Network Conf., Vol. 1, Paris, France, 1990.
- F. Galton, "Personal Identification and Description," Nature, June 1888.
- S. Carey and R. Diamond, "From Piecemeal to Configurational Representation of Faces," Science, Vol. 195, 1977.
- A.L. Yuille, D.S. Cohen, and P.W. Hallinan, "Feature Extraction from Faces using Deformable Templates," Proc. Of CVPR, San Diego, Calif., 1989.
- M. Kirby and L. Sirovich, "Application of the Karhunen-Loeve Procedure for the Characterisation of Human Faces," IEEE PAMI, Vol. 12, No. 1, 1990.
- H. Wang and S.F. Chang, "A Highly Efficient System for Automatic Face Region Detection in MPEG Video," IEEE Transactions on Circuits and Systems for Video Technology, Vol. 7, No. 4, Aug. 1997.
- L. Lorente and L. Torres, "Face Recognition of Video Sequences in a MPEG-7 Context using a Global Eigen Approach," Int'l Conf. Image Processing Vol. 4, Asahi Kosoku Printing, Japan, 1999.
- D. Rowland et al., "Transforming Facial Images in 2 and 3-D," Proc. Imagina 97 Conferences, 1997.
- B. Moghaddam, T. Jebara, and A. Pentland, "Bayesian Face Recognition," Pattern Recognition, Vol. 33, No. 11, Nov. 2000.
- A. Giachetti, "Matching Techniques to Compute Image Motion," Image and Vision Computing, Vol. 18, Jun. 2000.
- R. Fisher, "Image Processing Teaching Materials," Univ. of Edinburgh, http://www.dai.ed.ac.uk/HIPR2 (Oct. 16, 2001).
- Pattern Recognition Using Neural Networks: Theory and Algorithms for engineers and Scientists, Carl G. Looney, 1997 Oxford University press.
- Image Processing: The Fundamentals, Petrou Maria, 1999, John Wiley.
- Introduction to Algorithms, Thomas H. Cromen, Charles E. Leiserson, Ronald. Rivest 1994, MIT press.
- Fundamentals of Artificial Neural Networks, Mohammad H Hassoun, 1995, MIT press.
- Theory and applications of Neural Networks, Taylor J.G. and Mannion C.L.T. 1990.
- Introduction to Neural & Cognitive Modelling, Levine D.S. 1991.
- Lecture notes from: Synchronous Concurrent Algorithms, Mathew J. Pool. University of Swansea. - 1999.
- Lecture notes from: Applications of Artificial Intelligence, Dr J. Grant. University of Swansea. - 1999.
- Lecture notes from: Neural Networks. Dr T. Windeatt. University of Surrey. - 1999.
- COX, J. I. Feature based face recognition using mixture distances. IEEE Press, 1996
- KOHONEN, T. The self organizing maps. Springer-Verlang, Berlin 1995.
- PARKER, J. R. Algorithms for Image Processing and Computer Vision. John Wiley & sons, New York 1997.
- T. Rowley, "Silicon Fingerprint Readers: A solid state approach to biometrics", Proc. of the CardTech/SecureTech, Orlando, Florida, May 97, Vol. 1
- B. Schneir, "Security pitfalls in cryptography", Proc. of CardTech/SecureTech, Washington D.C., April 98, Vol. 1, pp. 621-626.
- B. Schneier, "The uses and abuses of biometrics".Communications of the ACM, August 1999, Vol. 42,
- Adini, Y., Moses, Y., and Ullman, S. (1997) Face recognition: the problem of compensating for changes in illumination direction. Trans. Pat. Anal.
- Belhumeur, P.N., Hespanha, J.P., and Kriegman, D.J. (1997) Eigenfaces vs. Fisherfaces: Recognition using class-specifc linear projecti