`
`February 2000/Vol. 43, No. 2 COMMUNICATIONS OF THE ACM
`
`IPR2022-00602
`Apple EX1011 Page 1
`
`
`
`
`
`When it comes to working biometric identification technologies,
`it’s not only our fingerprints that do the talking. Now, our eyes, hands,
`signature, speech, and even facial temperature can ID us.
`
`BIOMETRIC
`IDENTIFICATION
`Q
`
`uestions related to the identity of individuals such as “Is this the
`
`person who he or she claims to be?,” “Has this applicant been here before?,”
`
`“Should this individual be given access to our system?” are asked millions of
`
`times every day by organizations in financial services, health care, e-commerce,
`
`telecommunication, and government. In fact, identity fraud in welfare disbursements, credit
`
`card transactions, cellular phone calls, and ATM withdrawals totals over $6 billion each year [5].
`
`For this reason, more and more organizations are
`looking to automated identity authentication sys-
`tems to improve customer satisfaction and operating
`efficiency as well as to save critical resources (see Fig-
`ure 1). Furthermore, as people become more
`connected electronically, the ability to
`achieve a highly accurate automatic personal
`identification system is substantially more
`critical [5].
`Personal identification is the process of
`associating a particular individual with an
`identity. Identification can be in the form of verifi-
`cation (also known as authentication), which entails
`authenticating a claimed identity (“Am I who I
`claim I am?”), or recognition (also known as identi-
`
`fication), which entails determining the identity of a
`given person from a database of persons known to
`the system (“Who am I?”). Knowledge-based and
`token-based automatic personal identification
`approaches have been the two traditional
`techniques widely used [8]. Token-based
`approaches use something you have to make
`a personal identification, such as a passport,
`driver’s license, ID card, credit card, or keys.
`Knowledge-based approaches use something
`you know to make a personal identification,
`such as a password or a personal identification num-
`ber (PIN). Since these traditional approaches are not
`based on any inherent attributes of an individual to
`make a personal identification, they suffer from the
`
`Anil Jain, Lin Hong, and Sharath Pankanti
`
`WALTER SIPSER
`
`COMMUNICATIONS OF THE ACM February 2000/Vol. 43, No. 2
`
`91
`
`IPR2022-00602
`Apple EX1011 Page 2
`
`
`
`Figure 1. Biometric applications.
`
`(a) National ID card
`
`(b) Smartcard
`
`(c) ATM transaction
`
`(d) Computer login
`
`obvious disadvantages: tokens may be lost, stolen,
`forgotten, or misplaced, and a PIN may be forgot-
`ten by a valid user or guessed by an impostor. (Sur-
`prisingly, approximately 25% of the people appear
`to write their PIN on their ATM card, thus defeat-
`ing the protection offered by PIN when ATM
`cards are stolen [5]!) Because knowledge-based and
`token-based approaches are unable to differentiate
`between an authorized person and an impostor
`who fraudulently acquires the token or knowledge
`of the authorized person [8], they are unsatisfac-
`tory means of achieving the security requirements
`of our electronically interconnected information
`society.
`Biometric identification refers to identifying an
`individual based on his or her distinguishing physi-
`ological and/or behavioral characteristics (biometric
`identifiers) [5]. It associates/disassociates an individ-
`ual with a previously determined identity/identities
`based on how one is or what one does. Because
`many physiological or behavioral characteristics are
`distinctive to each person, biometric identifiers are
`inherently more reliable and more capable than
`knowledge-based and token-based techniques in dif-
`ferentiating between an authorized person and a
`fraudulent impostor.
`A biometric system is essentially a pattern recog-
`nition system that makes a personal identification
`by establishing the authenticity of a specific physio-
`logical or behavioral characteristic possessed by the
`user. Logically, a biometric system can be divided
`into the enrollment module and the identification
`module (see Figure 2). During the enrollment
`phase, the biometric characteristic of an individual
`is first scanned by a biometric sensor to acquire a
`digital representation of the characteristic. In order
`to facilitate matching and to reduce the storage
`
`92
`
`February 2000/Vol. 43, No. 2 COMMUNICATIONS OF THE ACM
`
`requirements, the digital representation is further
`processed by a feature extractor to generate a com-
`pact but expressive representation, called a “tem-
`plate.” Depending on the application, the template
`may be stored in the central database of the biomet-
`ric system or be recorded on a magnetic card or
`smartcard issued to the individual.
`During the recognition phase, the biometric
`reader captures the characteristic of the individual to
`be identified and converts it to a digital format,
`which is further processed by the feature extractor to
`produce the same representation as the template.
`The resulting representation is fed to the feature
`matcher that compares it against the template(s) to
`establish the identity of the individual.
`An ideal biometric should be universal, where
`each person possesses the characteristic; unique,
`where no two persons should share the characteris-
`tic; permanent, where the characteristic should nei-
`ther change nor be alterable; and collectable, where
`the characteristic is readily presentable to a sensor
`and is easily quantifiable.
`In practice, however, a characteristic that satisfies
`all these requirements may not always be feasible for
`a useful biometric system. The designer of a practi-
`
`Table 1. Biometric applications
`
`Forensic
`Criminal investigation
`Corpse identification
`Parenthood
` determination
`
`Civilian
`National ID
`Driver's license
`Welfare disbursement
`
`Commercial
`ATM
`Credit card
`Cellular phone
`
`Border crossing
`
`Access control
`
`cal biometric system must also consider a number of
`other issues, including:
`
`•Performance, that is, a system’s accuracy, speed,
`robustness, as well as its resource requirements,
`and operational or environmental factors that
`affect its accuracy and speed;
`•Acceptability, or the extent people are willing to
`accept for a particular biometric identifier in
`their daily lives;
`•Circumvention, as in how easy it is to fool the sys-
`tem through fraudulent methods.
`
`Depending on the application context, a biometric
`system may either operate in a verification (authen-
`tication) mode or in a recognition (identification)
`mode [5]. A verification system authenticates a per-
`son’s identity by comparing the captured biometric
`
`CARD, IRIS, AND THERMOGRAM, RESPECTIVELY
`AND MIKOS LTD. FOR PROVIDING THE PICTURES OF RETINA, SMART-
`THE AUTHORS ARE GRATEFUL TO EYEDENTIFY CORP., IRISCAN INC.,
`
`
`
`IPR2022-00602
`Apple EX1011 Page 3
`
`
`
`
`
`characteristic with the person’s own biometric tem-
`plate(s) prestored in the database. In this system, an
`individual who desires to be identified submits a
`claim to an identity usually via a magnetic-stripe
`card, login name, or smartcard, and the system
`either rejects or accepts the submitted claim of iden-
`tity. In a recognition system, the system establishes a
`subject’s identity (or fails to if the subject is not
`
`Figure 2. A generic biometric system.
`
`Enrollment
`
`Biometric
`Sensor
`
`Feature Extractor
`
`Identification
`
`Feature Extractor
`
`Biometric
`Sensor
`
`Feature Matcher
`
`enrolled in the system database) by searching the
`entire template database for a match—-without the
`subject having to claim an identity.
`
`Measuring Performance
`Evaluating the performance of a biometric identifica-
`tion system is a challenging research topic [12]. The
`overall performance of a biometric system is assessed in
`terms of its accuracy, speed, and storage. Several other
`factors, like cost and ease-of-use, also affect efficacy.
`Biometric systems are not perfect, and will some-
`times mistakenly accept an impostor as a valid indi-
`vidual (a false match) or conversely, reject a valid
`individual (a false nonmatch). The probability of
`committing these two types of errors are termed false
`nonmatch rate (FNR) and false match rate (FMR);
`the magnitudes of these errors depend upon how lib-
`erally or conservatively the biometric system oper-
`ates. Figure 3 shows the trade-off between a system’s
`FMR and FNR at different operating points; it’s
`called the “Receiver Operating Characteristics
`(ROC)” and is a comprehensive measure of the sys-
`tem accuracy in a given test environment.
`High-security access applications, where concern
`about break-in is great, operate at a small FMR.
`Forensic applications, where the desire to catch a
`criminal outweighs the inconvenience of examining a
`large number of falsely accused individuals, operate
`their matcher at a high FMR. Civilian applications
`attempt to operate their matchers at the operating
`
`Template Database
`
`points with both a low FNR and a low FMR. The
`error rate of the system at an operating point where
`FMR equals FNR is called the equal error rate (EER)
`which may often be used as a terse descriptor of sys-
`tem accuracy. Accuracy performance of a biometrics
`system is considered acceptable if the risks (benefits)
`associated with the errors in the decision-making at a
`given operating point on ROC for the given test envi-
`ronment are acceptable. Simi-
`larly,
`accuracy
`of
`a
`biometrics-based identification
`is unacceptable/poor if the risks
`(benefits) associated with errors
`related to any operating point
`on the ROC for a given test
`environment are unacceptable
`(insufficient).
`The size of a template, the
`number of templates stored per
`individual, and the availability
`of compression mechanisms
`determine the storage required
`per user. When template sizes
`are large and the templates are
`stored in a central database, network bandwidth may
`become a system bottleneck for identification. A typ-
`ical smartcard may only hold a few kilobytes of infor-
`mation (for instance, 8K) and in systems using
`smartcards to distribute the template storage, tem-
`plate size becomes an important design issue.
`The time required by a biometric system to make
`an identification decision is critical to many applica-
`tions. For a typical access-control application, the sys-
`tem needs to make an authentication decision in
`real-time. In an ATM application, for instance, it is
`desirable to accomplish the authentication within
`about one second. For forensic applications, however,
`the time requirements may not be very stringent.
`All other factors remaining identical, the wide-
`spread use of biometrics will be stimulated by its
`adoption in the consumer market. The single most
`important factor affecting this realization is the cost
`of the biometrics systems including the sensors and
`related infrastructure. Some sensors, such as micro-
`phones, are already very inexpensive, while others,
`such as CCD cameras, are now becoming standard
`peripherals in a personal computing environment.
`With the recent advances in solid-state technology,
`fingerprint sensors will become sufficiently inexpen-
`sive in the next few years. Storage requirements of
`the biometric templates and processing requirements
`for matching are among the two major considera-
`tions towards the infrastructure cost.
`The human factors issue is also important to the
`
`COMMUNICATIONS OF THE ACM February 2000/Vol. 43, No. 2
`
`93
`
`IPR2022-00602
`Apple EX1011 Page 4
`
`
`
`untary basis with either explicit or implicit incen-
`tives for opting biometrics-based solution.
`
`Applications Flourish
`Biometrics is a rapidly evolving technology that has
`been widely used in forensics, such as criminal iden-
`tification and prison security. Biometric identifica-
`tion is also under serious consideration for adoption
`in a broad range of civilian applications. E-com-
`merce and e-banking are two of the most important
`application areas due to the rapid progress in elec-
`tronic transactions. These applications include elec-
`tronic fund transfers, ATM security, check cashing,
`credit card security, smartcards security, and online
`transactions. There are currently several large bio-
`metric security projects in these areas under devel-
`opment,
`including
`credit
`card
`security
`(MasterCard) and smartcard security (IBM and
`American Express). A variety of biometric technolo-
`gies are now competing to demonstrate their effi-
`cacy in these areas.
`The market of physical access control is currently
`dominated by token-based technology. However, it
`is predicted that, with the progress in biometric
`technology, market share will increasingly shift to
`biometric techniques.
`Information system and computer-network secu-
`rity, such as user authentication and access to data-
`bases via remote
`login
`is another potential
`application area. It is expected that more and more
`information systems and computer-networks will be
`secured with biometrics with the rapid expansion of
`Internet and intranet. With the introduction of bio-
`metrics, government benefits distribution programs
`such as welfare disbursements will experience sub-
`stantial savings in deterring multiple claimants. In
`addition, customs and immigration initiatives such
`as INS Passenger Accelerated Service System
`(INSPASS), which permits faster processing of pas-
`sengers at immigration checkpoints based on hand
`geometry, will greatly increase the operational effi-
`ciency. A biometric-based national identification sys-
`tem provides a unique ID to the citizens and
`integrates different government services. Biometrics-
`based voter registration prevents voter fraud; and
`biometrics-based driver registration enforces issuing
`only a single driver license to a person; and biomet-
`rics-based time/attendance monitoring systems pre-
`vent abuses of the current token-based manual
`systems.
`
`Biometric Technologies
`There are a multitude of biometric techniques either
`widely used or under investigation. These include,
`
`Figure 3. Receiver Operating Characteristics
`(ROC) of a system illustrates false nonmatch
`rate (FNR) and false match rate (FMR) of a
`matcher at all operating points. Each point on
`a ROC defines FNR and FMR for a given
`matcher, operating at a particular matching
`score threshold. A smaller FNR (that is, a more
`tolerant system) usually leads to a larger FMR
`while a smaller FMR (a less tolerant system)
`usually implies a larger FNR. Note that System
`A is consistently inferior to System B in
`accuracy performance.
`
`Forensic
`Applications
`
`Equal Error Rate
`
`System A
`
`Civilian
`Applications
`
`System B
`
`High Security Access
`Applications
`
`False Nonmatch Rate
`
`False Match Rate
`
`success of a biometric-based identification. How
`easy and comfortable is it to acquire a given biomet-
`ric? For example, biometric measurements that do
`not involve touching an individual, such as face,
`voice, or iris, may be perceived as more user-friendly.
`Additionally, biometric technologies requiring very
`little cooperation/participation from the users (such
`as face and thermograms) may be perceived as more
`convenient to users. A related issue is public accep-
`tance. There may be a prevalent perception that bio-
`metrics are a threat to the privacy of an individual.
`In this regard, the public needs to learn that bio-
`metrics could be one of the most effective, and in
`the long run, more profitable means for protecting
`individual privacy. For instance, a biometrics-based
`patient information system can reliably ensure that
`medical records can only be accessed by medical per-
`sonnel and the individual concerned. As in any
`industry, government regulations and directives may
`either provide a boost or lead to the demise of cer-
`tain types of biometric technologies. Upcoming
`U.S. legislation such as the Health Information
`Portability Act (HIPA), may have a favorable impact
`on the biometrics industry. A good approach to
`piloting and gaining gradual acceptance of a bio-
`metrics solution could be to introduce it on a vol-
`
`94
`
`February 2000/Vol. 43, No. 2 COMMUNICATIONS OF THE ACM
`
`
`
`IPR2022-00602
`Apple EX1011 Page 5
`
`
`
`PERMISSION FROM KLUWER ACADEMIC PUBLISHING.
`THE PICTURES OF RETINA, SMARTCARD, IRIS, AND THERMOGRAM, RESPECTIVELY; FROM [5] USED WITH
`THE AUTHORS ARE GRATEFUL TO EYEDENTIFY CORP., IRISCAN INC., AND MIKOS LTD. FOR PROVIDING
`
`
`
`facial imaging (both optical and
`infrared), hand and finger geometry,
`eye-based methods (iris and retina),
`signature, voice, vein geometry, key-
`stroke, and finger- and palm-print
`imaging. Some of these methods are
`indicated in Figure 4.
`
`Figure 4. Examples of different biometric characteristics.
`
`face
`
`retinal scan
`
`Face. Facial images are probably the
`most common biometric character-
`istic used by humans to make a per-
`sonal identification. Identification
`based on face is one of the most
`active areas of research, with applica-
`tions ranging from the static, con-
`trolled mug-shot verification to a
`dynamic, uncontrolled face identifi-
`cation in a cluttered background [2].
`Approaches to face recognition are
`typically based on location and
`shape of facial attributes, such as the
`eyes, eyebrows, nose, lips, and chin
`shape and their spatial relationships;
`the overall (global) analysis of the
`face image and its break-down into a
`number of canonical faces, or a combination thereof.
`While performance of the systems [1] commer-
`cially available is reasonable, it is questionable
`whether the face itself, without any contextual infor-
`mation, is a sufficient basis for recognizing a person
`from a large number of identities with an extremely
`high level of confidence. It is difficult to recognize a
`face from images captured from two drastically dif-
`ferent views. Further, current face recognition sys-
`tems impose a number of restrictions on how the
`facial images are obtained, sometimes requiring a
`simple background or special illumination. In order
`for the face recognition systems to be widely
`adopted, they should automatically detect whether a
`face is present in the acquired image; locate the face
`if there is one; and recognize the face from a general
`viewpoint.
`
`facial thermogram
`
`fingerprint
`
`iris
`
`hand geometry
`
`signature
`
`voice print
`
`gram is a nonintrusive biometric technique which
`can verify an identity without contact. The claimed
`superiority of face thermogram-based recognition
`over visual face recognition using CCD cameras is
`based on the following observations: An infrared
`camera can capture the face thermogram in very low
`ambient light or in the absence of any light at all; the
`vascular structure may be more rich in information
`and remains invariant to intentional or uninten-
`tional variations in visual facial appearance [11].
`Although it may be true that face thermograms are
`unique to each individual, it has not been proven that
`face thermograms are sufficiently discriminative. Face
`thermograms may depend heavily on a number of
`factors such as the emotional state of the subjects, or
`body temperature, and like face recognition, face
`thermogram recognition is view-dependent.
`
`Facial Thermogram. The underlying vascular sys-
`tem in the human face produces a unique facial sig-
`nature when heat passes through the facial tissue and
`is emitted from the skin [11]. Such facial signatures
`can be captured using an infrared camera, resulting
`in an image called a “face thermogram.” It is claimed
`that a face thermogram is unique to each individual
`and is not vulnerable to disguises. Even plastic
`surgery, which does not reroute the flow of blood
`through the veins, is believed to have no effect on
`the formation of the face thermogram. Face thermo-
`
`Fingerprints. Humans have used fingerprints for
`personal identification for centuries and the validity
`of fingerprint identification has been well-estab-
`lished [6]. A fingerprint is the pattern of ridges and
`furrows on the surface of a fingertip, the formation
`of which is determined during the fetal period. They
`are so distinct that even fingerprints of identical
`twins are different as are the prints on each finger of
`the same person.
`With the development of solid-state sensors, the
`marginal cost of incorporating a fingerprint-based
`
`COMMUNICATIONS OF THE ACM February 2000/Vol. 43, No. 2
`
`95
`
`IPR2022-00602
`Apple EX1011 Page 6
`
`
`
`
`
`gerprint identification generally requires a large
`amount of computational resources. Finally, finger-
`prints of a small fraction of a population may be
`unsuitable for automatic identification because of
`genetic, aging, environmental, or occupational rea-
`sons.
`
`biometric system may soon become affordable in
`many applications. Consequently, fingerprints are
`expected to lead the biometric applications in the near
`future, with multiple fingerprints providing sufficient
`information to allow for large-scale recognition
`involving millions of identities. One problem with
`fingerprint technology is its lack of acceptability by a
`Hand geometry. A variety of measurements of the
`typical user, because fingerprints have traditionally
`human hand, including its shape, and lengths and
`been associated with criminal investigations and
`widths of the fingers, can be used as biometric char-
`police work. Another problem is that automatic fin-
`A Case Study in Biometrics
`T wo primary components of a
`
`compared from their original rep-
`resentations as the sensed fingers
`may be differently aligned with
`respect to the imaging system.
`The feature vectors are typically
`aligned based on some landmark
`information in the feature vector.
`In figures d, e, f, the properties
`of the ridge associated with
`minutiae are used to align the
`
`feature vectors. Once the feature
`vectors are aligned and overlaid,
`the number of corresponding
`minutiae, that is, minutiae in
`close proximity to each other with
`similar attributes, constitutes a
`basis for quantifying the likeli-
`hood of fingerprint feature vec-
`tors originating from the same
`finger.
`
`biometric-based identification
`system are the feature extractor
`and matcher. Here, we summarize
`typical steps involved in these two
`components for fingerprint-based
`authentication systems.
`The unprocessed input gray
`values of the fingerprint images
`are not invariant over the time of
`capture and are suscepti-
`ble to noise. Therefore,
`landmark features on a
`finger, for example, the
`fingerprint ridge endings
`and
`ridge bifurcations
`(collectively known as
`“minutiae”), are used in a
`fingerprint-based authen-
`tication system. The fea-
`ture extraction system
`detects the minutiae from
`the input image through a
`series of image processing
`steps (see figure). The fea-
`ture vector typically con-
`sists of a
`list of the
`locations and other attrib-
`utes (for example, orien-
`tation of the ridge) of the
`minutiae detected in a fin-
`gerprint image.
`A fingerprint matcher
`(see figures d, e, f) takes
`two feature vectors and
`determines whether the
`minutiae in the feature
`vectors originate from the
`same finger. The feature
`vectors cannot be directly
`
`Steps in fingerprint-based identification: (a) input fingerprint image;
`(b) orientation estimation for input image; (c) thinned ridges for input
`image; (d) input minutiae set overlaid on the input image; (e) template
`minutiae set overlaid on the template fingerprint image; and (f) matching
`result where template minutiae and their correspondences are connected
`by red lines. Matching score for this pair of input and template fingerprints
`was 630. The maximum matching score is 1,000 and the minimum
`threshold score for a pair to be considered as a valid match for a
`typical application using this matcher is 150.
`
`96
`
`February 2000/Vol. 43, No. 2 COMMUNICATIONS OF THE ACM
`
`IPR2022-00602
`Apple EX1011 Page 7
`
`
`
`
`
`Table 2. Comparison of biometric technologies based on perceptions of three biometrics experts [5].
`
`Biometrics
`Face
`Fingerprint
`Hand Geometry
`Iris
`Retinal Scan
`Signature
`Voice Print
`F. Thermogram
`
`Universality
`high
`medium
`medium
`high
`high
`low
`medium
`high
`
`Uniqueness
`low
`high
`medium
`high
`high
`low
`low
`high
`
`Permanence
`medium
`high
`medium
`high
`medium
`low
`low
`low
`
`Collectability
`high
`medium
`high
`medium
`low
`high
`medium
`high
`
`Performance
`low
`high
`medium
`high
`high
`low
`low
`medium
`
`Acceptability
`high
`medium
`medium
`low
`low
`high
`high
`high
`
`Circumvention
`low
`high
`medium
`high
`high
`low
`low
`high
`
`acteristics [9]. Hand geometry-based biometric sys-
`tems have been installed at hundreds of locations
`around the world. The technique is very simple, rel-
`atively easy to use, and inexpensive. Operational
`environmental factors such as dry weather, or indi-
`vidual anomalies such as dry skin, generally have no
`negative effects on identification accuracy. A main
`disadvantage of this technique is its low discrimina-
`tive capability. Hand geometry information may not
`be invariant over the lifespan of an individual, espe-
`cially during childhood. In addition, an individual’s
`jewelry or limitations in dexterity (for example, from
`arthritis), may pose further challenges in extracting
`the correct hand geometry information. Lastly,
`because the physical size of a hand geometry-based
`system is large, it cannot be used in certain applica-
`tions such as laptop computers.
`
`Retinal Pattern. The pattern formed by veins
`beneath the retinal surface in an eye is stable and
`unique [10] and is, therefore, an accurate and feasible
`characteristic for recognition. Digital images of reti-
`nal patterns can be acquired by projecting a low-
`intensity beam of visual or infrared light into the eye
`and capturing an image of the retina using optics
`similar to a retinascope. In order to acquire a fixed
`portion of the retinal vasculature needed for identifi-
`cation, the subject is required to closely gaze into an
`eye-piece and focus on a predetermined spot in the
`visual field. In many applications, the degree of user
`cooperation required in imaging a retina may not be
`acceptable to the subjects undergoing identification.
`Another disadvantage of this biometrics is that retinal
`scanners are expensive. A number of retinal scan-
`based biometric systems have been installed in several
`highly secure environments such as prisons.
`
`Iris. The iris is the annular region of the eye
`bounded by the pupil and the sclera (white of the
`eye) on either side. The visual texture of the iris sta-
`
`bilizes during the first two years of life and its com-
`plex structure carries very distinctive information
`useful for identification of individuals. Initial avail-
`able results on accuracy and speed of iris-based iden-
`tification are promising and point to the feasibility
`of a large-scale recognition using iris information.
`Each iris is unique and even irises of identical twins
`are different. Furthermore, the iris is more readily
`imaged than retina; it is extremely difficult to surgi-
`cally tamper iris texture information and it is easy to
`detect artificial irises (for example, designer contact
`lenses) [3]. Although the early iris-based identifica-
`tion systems required considerable user participation
`and were expensive, efforts are underway to build
`more user-friendly and cost-effective versions.
`It remains to be seen how this relatively recently
`discovered biometric matures and gains public
`acceptance.
`
`Signature. Each person has a unique style of hand-
`writing. However, no two signatures of a person are
`exactly identical; the variations from a typical signa-
`ture also depend upon the physical and emotional
`state of a person. The identification accuracy of sys-
`tems based on this highly behavioral biometric is
`reasonable but does not appear to be sufficiently
`high to lead to large-scale recognition. There are two
`approaches to identification based on signature [7]:
`static and dynamic. Static signature identification
`uses only the geometric (shape) features of a signa-
`ture, whereas dynamic (online) signature identifica-
`tion uses both the geometric (shape) features and the
`dynamic features such as acceleration, velocity, pres-
`sure, and trajectory profiles of the signature. An
`inherent advantage of a signature-based biometric
`system is that the signature has been established as
`an acceptable form of personal identification
`method and can be incorporated transparently into
`the existing business processes requiring signatures
`such as credit card transactions.
`
`COMMUNICATIONS OF THE ACM February 2000/Vol. 43, No. 2
`
`97
`
`IPR2022-00602
`Apple EX1011 Page 8
`
`
`
`
`
`Speech. Speech is a predominantly behavioral bio-
`metrics. The invariance in the individual charac-
`teristics of human speech is primarily due to
`relatively invariant shape/size of the appendages
`(vocal tracts, mouth, nasal cavities, lips) synthesiz-
`ing the sound [4]. Speech of a person is distinctive
`but may not contain sufficient invariant informa-
`tion to offer large-scale recognition. Speech-based
`verification could be based on either a text-depen-
`dent or a text-independent speech input. A text-
`dependent verification authenticates the identity
`of an individual based on the utterance of a fixed
`predetermined phrase. A text-independent verifi-
`cation verifies the identity of a speaker indepen-
`dent of the phrase, which is more difficult than a
`text-dependent verification but offers more pro-
`tection against fraud. Generally, people are willing
`to accept a speech-based biometric system. How-
`ever, speech-based features are sensitive to a num-
`ber of factors such as background noise as well as
`the emotional and physical state of the speaker.
`Speech-based authentication is currently restricted
`to low-security applications because of high vari-
`ability in an individual’s voice and poor accuracy
`
`98
`
`February 2000/Vol. 43, No. 2 COMMUNICATIONS OF THE ACM
`
`performance of a typical speech-based authentica-
`tion system.
`
`Conclusions
`Biometrics refers to automatic identification of a
`person based on his or her physiological or behav-
`ioral characteristics. It provides a better solution for
`the increased security requirements of our informa-
`tion society than traditional identification methods
`such as passwords and PINs. As biometric sensors
`become less expensive and miniaturized, and as the
`public realizes that biometrics is actually an effective
`strategy for protection of privacy and from fraud,
`this technology is likely to be used in almost every
`transaction needing authentication of personal
`c
`identity.
`
`References
`1. Biometrics Consortium homepage; www.biometrics.org.
`2. Chellappa, R., Wilson, C., and Sirohey, A. Human and machine
`recognition of faces: A survey. In Proceedings of the IEEE 83, 5 (1995)
`705–740.
`3. Daugman, J.G. High confidence visual recognition of persons by a test
`of statistical independence. IEEE Trans. Pattern Anal. and Machine
`Intell. 15, 11 (1993)1148–1161.
`4. Furui, S. Recent advances in speaker recognition. Pattern Recognition
`Letters 18 (1997) 859–872.
`5. Jain, A.K. Bolle, R. and Pankanti S. (eds.). Biometrics: Personal
`Identification in Networked Society. Kluwer, New York, 1999.
`6. Jain, A.K., Hong, L., Pankanti, S., and Bolle, R. An identity-
`authentication system using fingerprints. In Proceedings of the IEEE
`85, 9 (1997), 1365–1388.
`7. Nalwa, V. Automatic on-line signature verification. In Proceedings
`of the IEEE 85, 2 (1997), 213–239.
`8. Miller, B. Vital signs of identity. IEEE Spectrum 31, 2 (1994),
`22–30.
`9. Sidlauskas, D.R. 3D hand profile identification apparatus. U.S.
`Patent No. 4736203, 1988.
`10. Hill, R.B. Apparatus and method for identifying individuals
`through their retinal vasculature patterns. US Patent No. 4109237,
`1978.
`11. Prokoski, F.K. Disguise detection and identification using infrared
`imagery. In the Proceedings of SPIE, Optics, and Images in Law
`Enforcement II. A.S. Hecht, Ed. (Arlington, VA, May, 1982),
`27–31.
`12. Wayman, J.L. Error Rate Equations for the General Biometric Sys-
`tem. IEEE Robotics & Automation 6, 9 (Jan. 1999), 35–48.
`
`Anil Jain (Jain@cse.msu.edu) is a University Distinguished Profes-
`sor at the Department of Computer Science and Engineering at Michi-
`gan State University.
`Lin Hong (lin@faceit.com) is a research staff member at Visionics
`Corp., Jersey City, NJ.
`Sharath Pankanti (sharat@us.ibm.com) is a research staff
`member at IBM T. J. Watson Research Center, Hawthorne, NY.
`
`Permission to make digital or hard copies of all or part of this work for personal or
`classroom use is granted without fee provided that copies are not made or distrib-
`uted for profit or commercial advantage and that copies bear this notice and the full
`citation on the first page. To copy otherwise, to republish, to post on servers or to
`redistribute to lists, requires prior specific permission and/or a fee.
`
`© 2000 ACM 0002-0782/00/0200 $5.00
`
`IPR2022-00602
`Apple EX1011 Page 9
`
`