Pattern Recognition 36 (2003) 361 – 369
`
`www.elsevier.com/locate/patcog
`
`Innovations in (cid:30)ngerprint capture devices
`Xiongwu Xia∗, Lawrence O’Gorman
`
`Veridicom Inc. 31 Scotto Pl, Dayton, NJ 08810, USA
`
`Received 21 December 2001
`
`Abstract
`
`The image capture device plays a key role in (cid:30)ngerprint authentication. In recent years, we have seen remarkable innovations
`in these devices, which have reduced the size, lowered the price, and improved the performance. These new sensors have
`paved the way for deployment of (cid:30)ngerprint authentication beyond law enforcement applications to more widespread personal
`authentication. This paper provides an overview of (cid:30)ngerprint capture devices. Sensor issues and future trends are also
`discussed. ? 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.
`
`Keywords: Fingerprint capture device; Authentication; Optical sensor; Solid-state sensor
`
`1. Introduction
`
`Biometrics o6ers to replace or complement passwords to
`o6er a higher level of user convenience and network se-
`curity. Fingerprint, voice, iris, and face comprise the most
`prevalent modalities of e6ective biometrics for computer se-
`curity [1]. No matter what kind of biometric, there is always
`a biometric capture device, and it is the predominant factor
`of system price and veri(cid:30)cation performance for the com-
`plete system.
`Inking is the traditional ways of (cid:30)ngerprint capture, which
`has a long history and is still being used by some law en-
`forcement applications. But it is inconvenient and time con-
`suming due to the subsequent digitization. On the contrary,
`the live scan (cid:30)ngerprint device can capture a digital (cid:30)nger-
`print image in real time.
`There are three types of live scan devices: optical,
`solid-state and other [2]. Optical (cid:30)ngerprint capture de-
`vices use a light source and lens to image the (cid:30)nger-
`print. The image is captured by a CCD=CMOS camera.
`Solid-state sensors appeared on the market in the mid-1990s.
`
`∗ Corresponding author. Tel.: +1-732-829-2281; fax: +1-732-
`355-0106.
`E-mail address: sam@veridicom.com (X. Xia).
`
`These sensors comprise an array of sensing elements that
`image the (cid:30)ngerprint via di6erent technologies. Usually the
`solid-state sensors have on-chip A=D (analog to digital con-
`verter) so that a digital image can be generated. The third
`category, “other”, includes devices that employ ultrasonic
`means for image capture [3].
`One occasional problem in (cid:30)ngerprint systems is the poor
`image quality. Fingerprint quality not only varies widely, but
`also changes over time. Elderly persons or manual workers
`tend to have poorer (cid:30)ngerprints. Even the same (cid:30)nger can
`be di6erent due to skin condition, weather conditions and
`(cid:30)nger cuts [4]. Since the quality and condition of human (cid:30)n-
`gerprints are quite di6erent, the image capture devices play
`a crucial role yielding a correct result for the authentication
`system. The ability to capture dry, wet, or other poor quality
`(cid:30)ngerprints becomes critical in commercial systems. Actu-
`ally the image quality for the same person’s (cid:30)nger can be
`quite di6erent using di6erent devices. The paper will dis-
`cuss such issues and measurements for (cid:30)ngerprint sensors.
`Fig. 1 illustrates various (cid:30)nger conditions.
`Low price, small size, and high recognition performance
`are the three challenges that must be met to achieve large de-
`ployment of a (cid:30)ngerprint device. Recent innovations for (cid:30)n-
`gerprint capture devices have shown considerable progress
`toward lower cost, smaller size, and good recognition. It
`is these devices that have paved the way toward personal
`
`0031-3203/02/$22.00 ? 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.
`PII: S 0 0 3 1 - 3 2 0 3 ( 0 2 ) 0 0 0 3 6 - 5
`
`CARDWARE EXHIBIT 2026, Page 1 of 9
`SAMSUNG V. CARDWARE PGR2023-00012
`
`

`

`362
`
`X. Xia, L. O’Gorman / Pattern Recognition 36 (2003) 361 – 369
`
`Fig. 1. Images of various (cid:30)nger conditions: (a) normal (cid:30)nger; (b) dry (cid:30)nger; (c) wet (cid:30)nger; (d) poor quality (cid:30)nger.
`
`authentication, where a (cid:30)ngerprint veri(cid:30)cation device is
`practical for any user of the internet, and for that matter a
`garage door or soda machine.
`This paper examines innovations of (cid:30)ngerprint capture
`devices. Sensor issues are discussed in the next section.
`Section 3 describes the optical sensor. Section 4 describes
`the solid-state sensor. Summary and conclusions are in
`Section 5.
`
`2. Sensor issues
`
`Important factors to describe and compare (cid:30)ngerprint cap-
`ture devices can be categorized as: cost, performance, and
`size [5]. We describe these below.
`
`Cost: Cost is obviously an important factor. Fingerprint
`capture devices have fallen in price from about US$1500 to
`$30 since the early 1990s. We can expect further reduction
`of price with more technical innovations and with larger
`volume sales of the devices.
`Performance: The performance includes various factors
`such as image resolution, bit depth, image quality, image
`capture area, and sensor durability.
`
`is
`The FBI image resolution standard for (cid:30)ngerprint
`500 dpi (dots per inch) [6]. Most commercial devices meet
`this requirement, however, some have lower resolutions
`down to 250 dpi. It is debatable what minimum resolution
`is suJcient for the population of users, however, it is under-
`stood that lower resolution may result in diJculty resolving
`ridges from valleys in (cid:30)ngerprints of people with narrow
`ridge spacing, and for children. Furthermore, some fea-
`ture extraction algorithms require higher resolution, though
`specialized algorithms may deal with lower resolution
`well.
`The FBI standard for pixel bit depth is 8 bits, which yields
`256 levels of gray. Some sensors actually capture only 2
`or 3 bits of real (cid:30)ngerprint information and then scale it to
`8 bits. Thus the e6ective bit depth is 2–3, not 8. There is
`no de(cid:30)nitive study that shows how recognition performance
`decreases when bit depth is decreased. However, it is under-
`stood that some degree of bit depth above 1 bit is necessary
`for good performance of many feature extraction algorithms.
`Image quality is another key factor. Most sensors can
`handle “normal” (cid:30)nger quality well. But the ability to cap-
`ture dry, wet, or poor quality (cid:30)ngerprints is more impor-
`tant in commercial applications. Some solid-state sensors
`have locally adjustable gain control. It enables automatic
`
`CARDWARE EXHIBIT 2026, Page 2 of 9
`SAMSUNG V. CARDWARE PGR2023-00012
`
`

`

`X. Xia, L. O’Gorman / Pattern Recognition 36 (2003) 361 – 369
`
`363
`
`Table 1
`Comparison of (cid:30)ngerprint capture devices
`
`Company
`
`Technology
`
`Type
`
`Identicator
`Digital Persona
`SecuGen
`Ethentica
`Veridicom
`Authentec
`In(cid:30)neon
`Atmel
`Veridicom
`
`Optical
`Optical
`Optical
`Electro-optical
`Capacitive
`E-(cid:30)eld
`Capacitive
`Thermal
`Capacitive
`
`Touch
`Touch
`Touch
`Touch
`Touch
`Touch
`Touch
`Swipe
`Swipe
`
`Area
`(in)
`0:6 × 0:72
`0:7 × 0:7
`0:53 × 0:64
`0:56 × 0:76
`0:6 × 0:6
`0:51 × 0:51
`0:43 × 0:56
`0:55 × 0:06
`0:51 × 0:1
`
`adjustment of pixel gain to better image diJcult (cid:30)ngers
`and diJcult areas of a (cid:30)nger. This is discussed further in
`Section 4.
`The FBI standard for image capture area is 1 × 1 in. This
`is suJcient area even for very large (cid:30)ngerprints. The larger
`the area of the (cid:30)nger that is captured, the more ridges and
`valleys are captured and the more distinctive the (cid:30)ngerprint
`is. Especially, for recent (cid:30)ngerprint capture devices for per-
`sonal use, area is sacri(cid:30)ced to reduce cost and to (cid:30)t into
`a smaller footprint. There is a very real tradeo6 here be-
`tween size and resulting recognition rate: the smaller the (cid:30)n-
`gerprint area, the worse the recognition rate. An exception
`to this is the “swipe” sensor, which is much smaller than
`other sensors and requires the user to move (swipe) their
`(cid:30)ngerprint across its surface. This is described further in
`Section 4.
`impact-
`Sensor durability includes scratch-resistance,
`resistance,
`impermeability
`to
`liquids,
`resistance
`to
`greasy residue buildup (latent image), and resistance to
`electro-static discharge (ESD). Older optical sensors meet
`these criteria well, since the glass surface (platen) upon
`which the (cid:30)nger is placed is quite durable. They should
`just be cleaned periodically. More recent optical sensors
`have plastic platens and special coatings to enhance image
`quality, both of which are susceptible to scratching and
`dirt buildup. Durability is more challenging for solid-state
`sensors. The (cid:30)rst rule for an electronics engineer is to
`refrain from touching the silicon chip, however, (cid:30)nger-
`print chips must be touched. Therefore, they must resist
`damage due to scratching, tapping with a sharp object,
`liquid permeability (in particular sodium chloride, which
`is corrosive and present in (cid:30)nger oils), and ESD at least
`above 8 kV (an industry minimum for most consumer
`electronics).
`
`Size: The development of solid-state sensors brought the
`size reduction from what was brick-size for an optical device
`to postage size. Further integration on the chip has enabled
`even smaller system size with incorporation of A=D circuitry
`and bus circuitry onto the chip. The optical sensor has an in-
`
`System size
`
`Resolution
`[dpi]
`
`Bits=Pixel
`
`Cost
`
`small
`small
`small
`small
`small
`small
`small
`smaller
`smaller
`
`331
`300
`450
`400
`500
`250
`500
`500
`500
`
`8
`8
`8
`8
`8
`8
`8
`8
`8
`
`Low
`Low
`Low
`Low
`Low
`Low
`Low
`Lower
`Lower
`
`herent size limitation due to the required optical path length
`between platen and imager, and tradeo6 between small fo-
`cal length versus optical distortion from wide-angle lenses.
`However, optical readers have reduced substantially in size
`as well due to technological improvements as discussed in
`Section 3. For incorporation into a mobile device such as
`laptop computer or PDA, the chip has been favored so far
`due to its much thinner package. Optical and electro-optical
`devices as well as solid-state devices are small enough
`now to be incorporated into a keyboard or PCMCIA
`card [7].
`Other Issues: There are other issues that must be con-
`sidered when evaluating sensors. These include power con-
`sumption, I=O interface (USB, parallel port), operation en-
`vironment (temperature, humidity), weight, and user popu-
`lation etc. Live (cid:30)nger detection can be a challenge task for
`sensors. Most sensors only take three dimension image and
`cannot be spoofed by a printed image.
`
`It is best to perform a comparative test of devices in a
`similar environment to determine the most appropriate de-
`vice for particular usage. Sensors are also designed to work
`best with the speci(cid:30)c (cid:30)ngerprint matching algorithms. For
`instance, some matchers may work well for a lower resolu-
`tion image while others may not. Software is also used to
`(cid:30)x common hardware drawbacks such as latent (cid:30)ngerprints.
`So there is no absolute comparison rule for sensors and we
`have to evaluate sensor and software as a whole system.
`Table 1 summarizes features of some (cid:30)ngerprint capture de-
`vices that are commercially available [8].
`
`3. Optical sensor
`
`Before the introduction of optical sensors, (cid:30)ngerprints
`were mainly captured for law enforcement applications.
`In a traditional automatic (cid:30)ngerprint identi(cid:30)cation system
`(AFIS), a (cid:30)nger is inked, rolled onto paper, and digitized
`by a scanner. This system is expensive and time consum-
`ing, not to mention messy. The development of live scan
`
`CARDWARE EXHIBIT 2026, Page 3 of 9
`SAMSUNG V. CARDWARE PGR2023-00012
`
`

`

`364
`
`X. Xia, L. O’Gorman / Pattern Recognition 36 (2003) 361 – 369
`
`compact sensor.
`S = u + v;
`
`1=u + 1=v = 1=f;
`
`u=v = d1=d0;
`where u is the optical distance from the (cid:30)nger to the lens, v
`is the optical distance from the sensor array to the lens, f
`is the focal length of the lens, d1 is the (cid:30)nger width and d0
`is the camera array width.
`The detector array of the CCD or CMOS camera is much
`smaller than the size of a (cid:30)ngerprint. Therefore, a lens serves
`the function of reducing magni(cid:30)cation (focusing) by having
`the object distance larger than the image distance. A smaller
`focal length lens requires smaller optical path to image the
`same object and thus smaller package. However, the tradeo6
`for a smaller package is possible image distortion since the
`distance between the center of a (cid:30)ngerprint to the lens and
`the edge of the image to lens is relatively much di6erent.
`The pre-molded plastic, modular design of core optical
`components has been a large contributor to less expensive
`optical sensors. These optical parts are now uniformly pre-
`cise and easy to produce in mass production. In addition,
`these parts are much more durable and never need cali-
`bration as did earlier optical designs. Since earlier designs
`were hand calibrated, there was the possibility that even the
`same (cid:30)nger has di6erent image quality with the same type
`but di6erent sensors [8,13]. Any inability to capture consis-
`tent quality images can signi(cid:30)cantly lower the reliability of
`recognition since (cid:30)ngerprint is often captured with one de-
`vice and veri(cid:30)ed on another. Modular design has solved the
`problem.
`Older optical sensors had a larger size because of the
`requirement that the length of the optical path from platen to
`lens be much longer than the size of the surface imaged. An
`alternative is to use multiple mirrors to maintain the optical
`path length, but to do so in a smaller package. The downside
`to this is that mirrors make the design more complicated.
`Usually one mirror is used in commercial scanners.
`The quality of optical components also inNuences (cid:30)nger-
`print image quality. Non-uniform parallel light source may
`cause the image brightness to be di6erent in various image
`areas. Film planar light is used to create a uniform and par-
`allel source light. As well, a prism protecting coat (such as
`silicone) can enhance the image, tolerate skin oils, and even
`protect the prism from scratching [13].
`Although optical sensor has been reduced in size, it is still
`bulky in the context of micro-electronics. In the following
`section, we discuss some techniques to make the optical
`sensor even smaller.
`
`3.2. Small size optical sensor
`
`Various approaches have been developed to make smaller
`sized optical sensors. Sheet prism with a number of micro-
`prisms in the (cid:30)nger contact surface has been shown to reduce
`
`Fig. 2. Optical sensor.
`
`devices obviated the need for inking since (cid:30)ngerprint is
`directly scanned.
`
`3.1. Tradition optical sensor
`
`Optical (cid:30)ngerprint capture is typically based on the frus-
`trated total internal reNection (FTIR) phenomenon, as illus-
`trated in Fig. 2 [9–11]. When a (cid:30)nger touches the platen,
`the refractive index is di6erent between the ridge and valley.
`The light that passes through the glass upon valleys (air on
`the glass surface) is totally reNected. The light that passes
`through the glass upon ridges is not reNected. The reNected
`light is focused by a lens onto a CCD or CMOS [12] cam-
`era where the image is captured. Because, FTIR images a
`three-dimensional surface, optical sensor is not deceived by
`presentation of a photograph or printed image of a (cid:30)nger-
`print.
`Several recent innovations have improved the optical sen-
`sor design:
`• CMOS camera versus CCD, the CMOS camera has an
`on-chip A=D that reduces the cost.
`• Modular optical core design, this is usually a pre-molded
`plastic lens unit
`that assures accuracy of the optical
`path without calibration that was required with previous
`glass-lens designs.
`• Improvements in optical components, such as (cid:30)lm planar
`light, and prism protecting coat, as are described below.
`
`CCD usually generates an analog video output, thus it needs
`a frame grabber card to convert the signal to a digital image.
`This makes it expensive and complicated for installation and
`maintenance. CMOS camera incorporates A=D conversion
`on-chip and lowers the cost and complexity of the system
`signi(cid:30)cantly.
`Optical path, S, is de(cid:30)ned as the total optical length be-
`tween the (cid:30)nger surface and sensor array. Since the (cid:30)nger-
`print size is (cid:30)xed (a typical design has the (cid:30)nger 15 mm in
`width and 20 mm in height), S can be determined by the
`lens focus and camera array size. A smaller S means a more
`
`CARDWARE EXHIBIT 2026, Page 4 of 9
`SAMSUNG V. CARDWARE PGR2023-00012
`
`

`

`X. Xia, L. O’Gorman / Pattern Recognition 36 (2003) 361 – 369
`
`365
`
`Fig. 3. Optical sensor with a sheet prism.
`
`the size [15,16]. Fiber optics is proposed to provide opti-
`mum optical path [10,14]. A two-dimensional photo-electric
`image sensor is deployed to capture the image [8,10].
`
`3.2.1. Optical sensor using a sheet prism
`Traditional optical sensors have a single large prism. The
`sheet prism has been developed to reduce size, which has
`a Nat surface and a number of “prismlets” adjacent to each
`other [16]. Each prismlet has an entrance surface and an exit
`surface, shown in Fig. 3. The width of the sheet is more than
`10 times the maximum thickness of any one of the prismlets.
`Since the sheet prism is Nat, the overall sensor can be very
`thin comparing with a traditional sensor.
`This sensor also operates on the principle of FTIR. Al-
`though the prism size can be reduced, the optical path re-
`mains the same. The sensor has to use several mirrors to
`reduce space, at the cost of added manufacturing and adjust-
`ment complexity. A clever improvement is to include two
`sheet prisms stacked together [16]. It will increase the opti-
`cal path in a limited space and reduce the image distortion.
`The sheet prism is smaller and less expensive than a
`single prism. It gives the sensor an opportunity to achieve
`advantage with cost and size.
`
`3.2.2. Electro-optical sensor
`Optical and electronic components can be combined
`to create an e6ective sensor [8,10,17]. The design has
`a two-dimensional photo-electric image sensor, transpar-
`ent support and light-emitting component. Because, the
`photo-image sensor is the same size of (cid:30)nger, there is no
`need for a lens to generate a smaller image and the optical
`path can be very short. This enables it to be almost as thin
`as a solid-state sensor.
`layers, as shown in
`The sensor consists of several
`Fig. 4. The transparent layer has an outer surface upon which
`the (cid:30)nger is applied. On the inside is a two-dimensional
`matrix of photo-electric elements separated by strip-shaped
`gaps. The light-emitting layer emits the light through the
`strip-shaped gaps, and this passes through the transparent
`layer to ridges or valleys of the (cid:30)ngerprint. Light is re-
`
`Fig. 4. Electro-optical sensor.
`
`Nected back at the valleys to the photo-image layer. Since
`the refractive index of (cid:30)nger and transparent layer are de-
`signed to be very close, the ridges will absorb light. The
`photo-electric elements are protected from light coming
`from the sources so as to deliver an output signal only in re-
`sponse to light that has been reNected towards photo-electric
`elements. Thus, the pattern of ridges and valleys will be
`generated to form a (cid:30)ngerprint image.
`The photo-image layer can be made up of light-sensitive
`TFT (thin-(cid:30)lm transistor) or a bundle of optical (cid:30)bers [10].
`The optical (cid:30)ber may improve the image contrast since it
`has a better optical path means.
`Instead of a light-emitting layer, a light-emitting polymer
`may be used for this type of sensor [8]. The polymer is
`inexpensive and can be embedded in other materials (e.g.,
`monitor glass, laptop screen, mouse, PC card, keyboard,
`smart card, etc.). The size of touch area on the glass surface
`can be made large without the same penalty as size has
`for solid-state devices. Besides size, this polymer will have
`di6erent characteristics (better or worse) of durability than
`optical or solid-state devices.
`
`4. Solid-state sensor
`
`Although, solid-state sensors (also called silicon or chip
`sensors) have been proposed in patent literature since the
`1980s, it was not until the middle 1990s that these have been
`commercially available. Solid-state sensors were designed
`to address many of the shortcomings of optical sensors at
`the time. Optical sensors were costly, bulky, and many pro-
`duced poor image quality due to dirt buildup or poor cali-
`bration. A distinct advantage of silicon sensors is the abil-
`ity to integrate additional functions onto the chip. These in-
`clude A=D conversion or integration of a processor core to
`perform all (cid:30)ngerprint feature extraction and matching on a
`single chip [18,19].
`There are mainly two types of solid-state sensors: capac-
`itive and temperature. Capacitive technology is the most
`prevalent [20,21]. It determines the distance to the (cid:30)nger-
`print ridges and (cid:30)ngerprint valleys by measuring the elec-
`tric (cid:30)eld strength, which drops o6 as the inverse of distance.
`Temperature sensitive sensors have been designed to image
`
`CARDWARE EXHIBIT 2026, Page 5 of 9
`SAMSUNG V. CARDWARE PGR2023-00012
`
`

`

`366
`
`X. Xia, L. O’Gorman / Pattern Recognition 36 (2003) 361 – 369
`
`The capacitance C is determined by
`
`C = k(s=d);
`
`where C is the capacitance, k is the dielectric constant, s
`is the surface area of the capacitor, and d is the distance
`between the electrodes of capacitor. It is also known that
`
`dQ=dt = Cdv=dt;
`
`where dQ=dt is the change of charge over time, and dv=dt
`is the voltage change over time.
`As illustrated in Fig. 5, since k and s are (cid:30)xed, the capac-
`itance C changes with d. Because Q can be set by charging
`the capacitor to a known value, the capacitor voltage v will
`change when C is changed due to the distance that each
`ridge (closer) or valley (further) is located from the capac-
`itor plate. Thus a (cid:30)ngerprint image can be determined by
`the measurement of the voltage output change over time at
`each capacitor of the sensor array.
`Since (cid:30)ngerprint conditions vary, one of the advantages
`of the capacitive sensor is being able to adjust the gain to
`ensure the best image quality. This can be done either by
`adjusting the amount of charge placed on the capacitor or the
`amount of time that the voltage discharges [24,25]. It can be
`incorporated into a feedback procedure whereby the image
`is captured at certain settings and its quality examined by
`software, then the settings changed toward those to improve
`the quality, and this process iterated until the image quality
`is best. This software-controlled automatic gain adjustment
`enables the sensor to handle a wider range of (cid:30)ngerprint
`image quality from wet to dry and from weak to strong.
`The image resolution is determined by the size of each
`capacitor plate. Ridges and valleys are typically 100–
`200 (cid:2)m in width for an adult. For capacitor spacing of
`50 (cid:2)m (500 dpi), this yields 2–4 pixels per ridge or valley
`width [26]. The solid-state sensor is covered by a layer of
`silicon dioxide several microns thick. This layer serves as
`the insulating layer for ESD, physical scratching and chem-
`ical permeation protection [27,28]. A block diagram of a
`capacitive sensor is shown in Fig. 6. It has a sensor array of
`300×300 in pixels and is fabricated from a standard CMOS
`process. The image capture area is 15 mm × 15 mm, and
`image resolution is 500 dpi. The sensor speed is 15 frames
`per second.
`
`4.2. Low-cost solid-state sensor
`
`Although the cost of solid-state sensors has brought (cid:30)n-
`gerprinting accessible for personal authentication applica-
`tions, products can never be too inexpensive. Solid-state
`sensor costs depend mainly on the area of the chip. A larger
`die costs more due to fewer dies per wafer and lower yield.
`The traditional touch sensor has a size of 15 mm × 15 mm
`since it has to cover the (cid:30)nger, which is large for a chip.
`One way to obtain an image from a smaller size sen-
`sor is for the user to swipe their (cid:30)nger across a smaller,
`linear sensor, and then piece together the “line” images
`
`Fig. 5. Capacitive sensor.
`
`the temperature di6erence of a (cid:30)nger related to touching
`ridges versus non-touching valleys [22,29].
`One challenge for silicon sensors is to sustain ESD with-
`out damage. There are a number of ways manufacturers have
`protected their sensors: grounded enclosure, grounded metal
`ring around the chip, grounded metal “plugs” within the sen-
`sor array, grounded metal mesh as a top chip layer, and a
`very thick protective surface coating. Another challenge is
`cost. Since the (cid:30)ngerprint is large and (cid:30)xed in size, chip
`designers cannot reduce chip size to lower cost per chip. In-
`stead, smaller integration enables further functionality to be
`placed on the same size sensor to reduce system cost. We
`discuss another means for lowering silicon sensor cost in
`Section 4.2.
`
`4.1. Capacitive sensor
`
`There are a number of di6erent proposed and commercial
`capacitive (cid:30)ngerprint sensors [20–24]. We describe below
`the general principle of how they work.
`The capacitive sensor has a sensing surface on which the
`(cid:30)nger is placed. Below this surface is a two-dimension array
`of capacitor plates. The array is large in number, typically
`300 × 300 = 90; 000 pixels and the capacitors are small,
`typically 50 (cid:2)m, so the entire array comprises the size of a
`(cid:30)ngerprint. The capacitors must be smaller than the width
`of the ridges and valleys to resolve these features. Capac-
`itors have two plates, one plate is built within the sensor,
`and the other plate is considered to be the skin of the (cid:30)nger-
`print. Capacitance varies as a function of the distance be-
`tween the plates, so the (cid:30)ngerprint ridges and valleys can be
`di6erentiated on the basis of their capacitive measurement
`as illustrated in Fig. 5. The capacitive sensor also cannot
`be deceived by presentation of a Nat photograph or printed
`image of a (cid:30)ngerprint since it measures the distances.
`
`CARDWARE EXHIBIT 2026, Page 6 of 9
`SAMSUNG V. CARDWARE PGR2023-00012
`
`

`

`X. Xia, L. O’Gorman / Pattern Recognition 36 (2003) 361 – 369
`
`367
`
` 300 * 300
`
`Sensor Array
`
`MUX, LOGIC
`
`8-bit A/D
`
`Register:
`
`Row Decode
`
`Col Decode
`
`Fig. 6. Capacitive sensor block diagram.
`
`[29–31]. The concept of swiping is widely used already.
`People swipe credit cards at grocery stores, key cards for
`entry to hotel rooms, and identity cards at company cafete-
`rias. The size of silicon sensor can be reduced by a factor of
`10 and the cost reduced commensurately. Furthermore, it is
`possible to capture a larger image than for touch sensors if
`the user swipes a longer area of the (cid:30)nger. The larger area
`will improve recognition.
`In practice, a swipe sensor is not a single imaging row.
`Reconstruction is accomplished by determining overlap
`between adjacent slices by correlation,
`therefore,
`there
`must be some number of rows. This number relates to the
`speed of swiping, the speed by which the data can be ac-
`cepted (via a bus or serial interface), and the reconstruction
`
`algorithm requirement on the amount of overlap. A 64-row
`capacitive sensor is published for the (cid:30)ngerprint capture
`[30]. The rows of the sensor can be reduced if the speed
`of the sensor can be increased. A thermal swipe sensor has
`been developed and it uses 8 rows. Since it measures tem-
`perature di6erences, the image is weakest when the sensor
`temperature is the same as the skin temperature. In Fig. 7,
`we illustrate a capacitive sensor that is 8 rows high, with
`2–3 rows minimum of overlap.
`It may be possible to construct a sensor even down to a
`single row if there is some means for measuring swipe speed.
`For instance, a mechanical roller may be rotated as the (cid:30)nger
`swipes across a single-row reader, with the rotation of the
`roller recording the speed [31].
`The moving action across sensor may be complex and
`inconsistent. In some cases, a user may have to learn how to
`swipe to get a proper image during the (cid:30)rst usage of swipe
`sensor. In an experiment we tested 35 people for swiping
`action. Most people can learn to become comfortable with
`the swiping action within 5 trials.
`The slice reconstruction algorithm plays a vital role in
`image quality for a swipe sensor. It will rely on the overlap
`between adjacent slices. The slice overlaps may change be-
`cause of di6erent swiping speed and it is not a problem for
`reconstruction. However, there is a maximum swipe speed
`limit due to the sensor acquisition speed. A good reconstruc-
`tion algorithm shall not sacri(cid:30)ce ease of use for the swipe
`action.
`Some advantages of a swipe sensor over a touch sensor
`are:
`• Much lower cost, 1
`5 – 110 of a touch sensor.
`
`• Very small size, which will allow it to (cid:30)t into small mobile
`devices such as cellular phones and PDAs.
`
`Fig. 7. Image reconstruction from a swipe sensor.
`
`CARDWARE EXHIBIT 2026, Page 7 of 9
`SAMSUNG V. CARDWARE PGR2023-00012
`
`

`

`X. Xia, L. O’Gorman / Pattern Recognition 36 (2003) 361 – 369
`368
`• Lower power consumption, which could be critical for
`handheld devices.
`• Better recognition performance when a longer length im-
`age is captured.
`• More durable due to smaller sensor area to damage via
`impact or ESD.
`• Self-cleaning, the swiping action cleans the device.
`• No latent image. A latent image can be left from the oil
`residue of a previously applied (cid:30)nger on a touch sensor.
`The swiping action leaves no more than a slice size of
`residue.
`
`Since swipe sensors are just emerging it is not clear whether
`touch or swipe or both types of sensors will gain customer
`acceptance. Despite the advantages listed above, the swipe
`sensor does require the user to perform an action, which
`some users may deem an ergonomic disadvantage.
`
`5. Summary and conclusions
`
`Various (cid:30)ngerprint capture devices have been discussed
`in this paper. Optical sensors and solid-state sensors de-
`scribed along with their underlying technology, advan-
`tages and disadvantages. The recent advances in cost,
`size, and performance have moved (cid:30)ngerprint capture de-
`vices from small-volume law enforcement applications to
`the larger-volume arena of personal authentication. With
`increasing awareness of the importance of security and
`privacy in today’s networked world, there is no doubt that
`(cid:30)ngerprint biometrics will be part of the personal authen-
`tication solution. The remaining technological advance is
`lower cost and smaller size. This will enable wide-ranging
`applications for (cid:30)ngerprint authentication as this solution
`approaches the prevalence of the metal key, but o6ers much
`higher security and convenience.
`
`References
`
`2000,
`
`imager based on a-Si:H active-matrix
`[7] J. Lan, Fingerprint
`photo-diode arrays, ethentica and tactilesense white paper,
`2000, http://www.ethentica.com.
`[8] L. O’Gorman, An overview of commercial biometric
`systems, Workshop on Automatic Identi(cid:30)cation Advanced
`Technologies, Long Island, NY, 1997, pp. 23–26.
`[9] M. Kawagoe, A. Tojo, Fingerprint pattern classi(cid:30)cation,
`Pattern Recognition 17 (3) (1984) 295–303.
`[10] I. Fujieda, Y. Ono, S. Sugama, Fingerprint image input device
`having an image sensor with openings, US Patent 5446290,
`1995.
`[11] R.D. Bahuguna, T. Corboline, Prism (cid:30)ngerprint sensor that
`uses a holographic element, Appl. Opt. 35 (26) (1996) 5242–
`5245.
`products,
`division
`image
`[12] STMicroelectronics
`http://www.vvl.co.uk/products/image sensors.
`[13] SecuGen white
`paper,
`2000,
`http://www.secugen.com/
`index2.html http://www.secugen.com/index2.html.
`[14] R.F. Dowling Jr., K.L. Knowlton, Fingerprint acquisition
`system with a (cid:30)ber optic block, US Patent 4785171, 1988.
`[15] W.S. Chen, C.L. Kuo, Apparatus for imaging (cid:30)ngerprint or
`topographic relief pattern on the surface of an object, US
`Patent 5448649, 1995.
`[16] G. Zhou, Y. Qiao, F. Mok, Fingerprint sensing system using
`a sheet prism, US Patent 5796858, 1998.
`[17] M. Calmel, Fingerprint sensor device, US Patent 6128399,
`2000.
`[18] S. Shigematsu, H. Morimura, Y. Tanabe, K. Machida,
`A Single-chip (cid:30)ngerprint sensor and identi(cid:30)er,
`IEEE J.
`Solid-State Circuits 34 (12) (1999) 1852–1859.
`[19] S.
`Jung, A Low-power and high-performance CMOS
`(cid:30)ngerprint
`sensing and encoding architecture,
`IEEE J.
`Solid-State Circuits 34 (7) (1999) 978–984.
`[20] C. Tsikos, Capacitive (cid:30)ngerprint sensor, US Patent 4353056,
`1982.
`[21] D.R. Setlak, Electric (cid:30)eld (cid:30)ngerprint sensor apparatus and
`related methods, US Patent 5963679, 1999.
`[22] D.G. Edwards, Fingerprint sensor, US Patent 4429413, 1984.
`[23] A.G. Knapp, Fin