All-in-Focus Iris Camera With a Great Capture Volume

Kunbo Zhanga, Zhenteng Shenb, Yunlong Wanga and Zhenan Sunaa Center for Research on Intelligent Perception and ComputingNational Lab of Pattern Recognition, Institute of Automation

Chinese Academy of Sciences, Beijing, Chinab Tianjin Academy for Intelligent Recognition Technologies, Tianjin, China

kunbo.zhang@ia.ac.cn, shenzhenteng@tj.ia.ac.cn, yunlong.wang@cripac.ia.ac.cn,

znsun@nlpr.ia.ac.cn

Abstract

Imaging volume of an iris recognition system has beenrestricting the throughput and cooperation convenience inbiometric applications. Numerous improvement trials arestill impractical to supersede the dominant fixed-focus lensin stand-off iris recognition due to incremental perfor-mance increase and complicated optical design. In thisstudy, we develop a novel all-in-focus iris imaging systemusing a focus-tunable lens and a 2D steering mirror togreatly extend capture volume by spatiotemporal multiplex-ing method. Our iris imaging depth of field extension systemrequires no mechanical motion and is capable to adjust thefocal plane at extremely high speed. In addition, the mo-torized reflection mirror adaptively steers the light beam toextend the horizontal and vertical field of views in an activemanner. The proposed all-in-focus iris camera increases thedepth of field up to 3.9 m which is a factor of 37.5 comparedwith conventional long focal lens. We also experimentallydemonstrate the capability of this 3D light beam steeringimaging system in real-time multi-person iris refocusing us-ing dynamic focal stacks and the potential of continuous irisrecognition for moving participants.

1. Introduction

The capture volume of an imaging system refers to thesize of space in which the scene points appear in acceptablysharp focus in visual information acquisition. Large cap-ture volume is a longstanding goal in many optical imagingrelated applications such as tomography, biometrics, andmicroscopy. Limited capture volume has been the bottle-neck of iris recognition throughput improvement comparedwith the rapid development in processing algorithm. In-

978-1-7281-9186-7/20/$31.00 ©2020 IEEE

Deformable Lens

Horizontal FoV

VerticalFoV

DoF

Original Imaging Volume

Extended Imaging Volume

Iris Imaging System

DoF Extension

FoV ExtensionSteering Mirror

Focus Distance

Figure 1. Capture volume extension in an iris imaging system.DoF extension allows a participant to stand at various distances.FoV expansion extends the capture capability of participants atvarious orientations with different heights.

creasing the range of depth of field (DoF) and field of view(FoV) will allow less cooperation from participants [17].To extend DoF, one can reduce the aperture size at the ex-pense of decrease in light collection and signal-to-noise ra-tio (SNR) [15]. Other capture volume extension trials havefocused on multiple cameras [14], camera rotation [24], andzoom lens [17] in iris imaging. Since there is a trade-offbetween resolution and capture volume as shown in Fig-ure 1, the dominant optical design is still fixed-focus cam-era in industry. Focal Sweep and multiplexing have beenproposed as techniques to extend the capture volume of animaging system while maintaining high resolution and fastresponse [4, 16].

Mechanical motion based focal sweep was proposed byZhou [26] using a micro-actuator for seamless refocusing.Following this idea, recent advances have been able to ex-tend the depth range on the order of 100’s of centimeter.Telescope imaging was introduced to achieve larger stand-off distance on the order of 1’s of meter in range distanceiris recognition [22]. Bulky optical and focus control com-ponents, however, implies critical application issues. Tele-photo zoom iris imaging is slow due to mechanical motion

arX

iv:2

011.

0990

8v1

[ee

ss.I

V]

19

Nov

202

0

and long travel distance with uncertain lifetime [13]. In thispaper, we seek to overcome such drawbacks by integratinga deformable liquid lens with existing system to demon-strate the advantages of focal sweep in DoF extension ofiris recognition. Such a bio-inspired imaging mechanism issimilar to human vision which changes the focal point usingciliary muscle to adjust lens curvature instead of moving theposition of lens [12].

FoV of a singlet or zoom lens is fixed and narrow intelephoto optical system. To extend the field of view, Fu-nahasahi [9] developed a pan-tilt-zoom (PTZ) iris imagingdevice to track human head motion in an expanded area. Asobject distance increases, the weight and travel distance oflens components slow the time response in dynamic scene.Digital micromirror device as a light and compact alterna-tive has been widely used in micro-scale imaging [5] to ob-serve wider field while maintaining high resolution. Evenoff-the-shelf MEMS mirror size is not sufficient for irisimaging, it is proven to capture temporal multiplexing par-allel videos of four different objects using a single camerafor almost real-time [4]. We focus our research on steer-ing mirror based FoV extension together with focus-tunablelens based DoF extension to expand iris capture volumeover all three dimensions.

The following are the major contributions of this paper.Focus-tunable lens for DoV extension. We present a

focus-tunable iris camera by integrating a deformable lenswith a telephoto zoom lens. This new imaging device isable to adjust the desired focal plane by electrically chang-ing optical power with superior dynamic response as shownin Figure 2. A comparable variable focal length iris imag-ing solution requires complicate design of lens groups andmechanical motion which introduce significant time delayand challenges in durability.

Adaptive 2D steering mirror for FoV extension. Theintroduction of an adjustable mirror allows the precise lightpath manipulation with fast response for temporal multi-plexing iris imaging. Instead of rotating the heavy tele-photo lens, the reflection mirror with gold-coated fine pro-cessed surface is agile in motion control. Literally this 2Dsteering mirror can capture a participant at any position in a360◦surrounding.

All-in-focus iris camera with a great capture volume.Previous long range iris imaging researches are mostly re-stricted by its extension capability. In this study, we demon-strate that our compact deformable lens increases the DoF37.5 times and the steering mirror expands the FoV to±180◦(H) and ±60◦(V) for stand-off iris recognition. Ourproposed all-in-focus iris imaging system supersedes tradi-tional capture volume extension methods such as PTZ andwavefront coding in deployment and user cooperation.

Multi-person iris refocusing. The great capture vol-ume of our proposed system extends application scenarios

2D Steering Mirror

Adapter

Telephoto ZoomLens

Motorized Pivot

Rotation Directions

Eyes

Image SensorFocus-tunable Lens

Figure 2. Overview of our all-in-focus iris imaging systemschematic. The telephoto zoom lens and image sensor are placedvertically downwards. The focus-tunable lens is mounted betweenthe zoom lens and the image sensor with a customized ring adapterto compensate back focal length. The reflection mirror is motor-ized to steer the light back from the eyes into the sensor by two-dimensional rotation.

of iris recognition. We demonstrate the usage of spatiotem-poral multiplexing in iris imaging by examples such as realtime multi-person auto-refocusing and continuous iris cap-ture for a moving object.

2. RELATED WORK

As one of the critical measures of iris recognition systemperformance, capture volume is the 3D volume in which theimaging device capable to capture sufficiently qualified eyeimages of a participant [17]. Larger capture volume in freespace enables less user cooperation and better recognitionperformance. The longitudinal and the transverse dimen-sions in capture volume are defined by DoF and FoV re-spectively. Most capture volume extension works concen-trate on extending DoF by engineering innovative opticalelements such as wavefront coding, focal sweep, telescopezoom lens, and light field camera [19, 25]. Traditional DoFextension approach is to increase F-number by using longerfocal length or smaller aperture.

The first known multi-biometric acquisition system witha 3-6 meters stand-off range uses the long focal zoom lensdesign to extend the focus point by mechanical adjustmentof group zoom lens as illustrated in [1]. Considering thebulky size and slow reaction, new computational imagingmethod is introduced for rapid biometric information acqui-sition [25]. A light-field camera uses computational basedauto-refocusing method for DoF extension to overcome thetrading-off between DoF and aperture size in a conventionalcamera. However, the micro-lens based light field imagingcomes at the cost of a sacrifice in spatial resolution and dif-ficulties in fabrication. Recent works by Dong et al. [8] andHsieh et al. [10] use wavefront coding and post-processing

to compensate the blurring caused by phase mask for DoFextension. However, necessary image restoration algorithmand cubic phase mask optimization increase the complexityof system implementation to maintain invariant point spreadfunction (PSF) [20]. The limited DoF extension capability(6cm to 18cm) makes such kind of computational imagingtechniques impractical in complex scenes [19].

Since longer focal lens covers a smaller view angle,methods such as pan-tilt-zoom (PTZ) and multiple cam-eras are proposed to expand the FoV in iris imaging [23].Venugopalan [22] uses an off-the-shelf PTZ camera withthe 360◦ pan and 180◦ tilt capabilities to capture uncon-strained iris image between 0.5 m and 1.6 m. Even the pan-tilt movement with an adjustable speed up to 450◦/second,the heavy and bulky optical form factors will jeopardize theperformance of response time and angular resolution whenit comes to distant object. Another intuitive solution to ex-pand the view angle of iris image capture is by increasingthe number of cameras as demonstrated in the well-knownIOM system from Sarnoff in 2006 [14]. In this IOM systemthe 20 cm capture height of a single camera is increased to37 cm and 70 cm by vertically stacking two cameras andfour cameras separately. If a camera matrix is used to coverlarge view angle, the geometry size and calibration workwill become an obstacle that cannot be ignored [3].

In contrast, our all-in-focus iris capture system can ex-tend DoF with much larger ranges (3.9m when focal planeis set at 5m) and 360◦ omni-directional view without par-ticipant height restriction by manipulating light beam direc-tion in three-dimensional space. To extend DoF, a focus-tunable liquid lens is used in this iris camera by sweepingfocal planes at extremely high speed without any mechani-cal motion and optical system modification. Miau [16] ex-tends the DoF of face video more than 10m at 20fps us-ing periodic focal stacks acquired through deformable lens.To extend the horizontal and vertical FoVs, we steer a fast2D reflection mirror instead of the heavy telephoto com-ponents. Optical multiplexing imaging has been success-fully demonstrated to expand FoV in microscopy imagingby controlling steering mirror to superimpose images frommultiple FoVs onto a single focal plane [21]. The major ad-vantage of this system design is the dramatically increasedcapture volume while maintaining high spatial, temporaland angular resolutions for distant iris imaging.

3. PROPOSED IRIS IMAGING SYSTEMTo design the hardware for a distant iris imaging system,

several fundamental components such as imaging sensor,lens, and illumination should be taken into consideration.In this section, the hardware configurations are discussed interms of evaluating the capability to extend iris capture vol-ume towards better iris image quality acquisition. The aver-age diameter of human irides is on the order of 1 cm. Even

the National Institute of Science and Technology (NIST)showed that 120 pixels across iris diameter is suitable forrecognition [18], we will still employ the 200 pixels acrossthe iris region from ISO standard as the threshold in thisstudy according to our research experience on long distancescenario [11] . We explain the design decisions leading usto build the major components in this prototype imagingsystem. And we conclude this section by describing howlarge the capture volume is expanded in 3D by combiningthe focus-tunable lens and 2D steering mirror in theoreticalcalculation.

In this paper, our prototye system uses an image sensorwith pixel numbers 4080× 3072 at 30.5 fps and 35% rela-tive response at 850nm. In order to investigate the capturevolume change, we designed and manufactured an infraredzoom lens to cover the distance between 1m and 5m witha variable focal length between 70mm and 350mm. Sucha 15-piece f / 4.8 lens has view angles of 18◦x18◦ at 70mmfocal length and 3.8◦x3.8◦at 350mm. According to the lensequation under paraxial approximation, the DoF of an imag-ing system can be estimated as:

DoF = DN +DF =2Cd

fP/(d− f)− C2(d− f)/fP(1)

where the near limit DN and far limit DF are a functionof circle of confusion C, subject to lens distance d, focallength f , and exit pupil P . When the focal plane is set at5m, the DoF can be derived as 91mm where f = 350mm,d = 5000mm and the F-number of the lens is 4.8. Hencethe capture volume is defined by the size of DoF x FoVwhich is approximately 0.04m3

Since the intricate iris texture is well captured at wave-length between 700 and 900 mm for Asian eyes, we havedesigned an adaptive illuminator to provide uniform cover-age over the entire frontal view between 1 m and 5 m for anyparticipant not higher than 2.2 m. This 850 mm illuminatoris designed as matrix of individually controlled illumina-tion modules which meet the Class I LED safety require-ments in IEC standard 60825-1. When the participant eyeis detected in the permissible space, only the illuminationmodule governing the specified volume is turned on with aduration time matching the sensor exposure time.

3.1. Focus-Tunable Lens for DoF Extension

Compared with a conventional singlet lens with fixedfocal distance, the focal length of a focus-tunable lens isadjustable dynamically. In our imaging system, we focuson deformable surface approach due to its large aperturesize and superior dynamic response in which the membraneshape is changed by applying an electric current. We choosethe Optotune tunable lens EL-16-TC-20D with a clear aper-ture of 16 mm. The major hallmark of its large aperture

S

Objective

fo ft(t)

Focus TunableLens Sensor

O dot dts

(b)(a)

Figure 3. The focus-tunable lens model (a) and the 3D lightbeam steering prototype system (b). We use a compound systemof two thin lens to model our imaging system. The main opticalaxis of the lens is aligned with the center of the reflection mirror.

benefits the collection of infrared light reflected back fromthe eyes of a participant at a long standoff distance [2].

The optical power of this focus-tunable lens is controlledby changing the electrical current flowing through the coilof the actuator over an optical power range of -10 dpt and+10 dpt corresponding to -100 mm and 100 mm within con-served temperature region. Both the optical fluid and themembrane material are highly transparent in the range of400 and 2500 nm which is ideal for iris imaging. The lenscontrol driver is able to set the current with a 0.07mA stepwith frequencies from 0.2 to 2000 Hz. The response andsettling times are 5 ms and 25 ms which are magnitude fasterthan most mechanical alternatives. We manufactured an ad-ditional spacer plus the general c-mount to attach this tun-able lens between our long focal lens and the image sensorwith optimal focusing result on imaging plane. In practice,we typically use 1mA as the minimal step size for focusadjustment which roughly equal to 1 cm in the depth direc-tion. Since the liquid lens repeatability error of ±0.1 dptdoes not have obvious effect on the quality of eye imagefor recognition, we did not employ any control strategy forcompensation.

To estimate the focal length of our optical components,we model our implementation as a compound system oftwo thin lens with one objective and one focus-tunable lens.Such an approximation is reasonable because most geomet-ric aberrations have been corrected in our designed zoomlens which is simple for modelling. It is required to apply alarge separation between the principal planes of the focus-tunable lens and the telephoto lens to minimize geometricaberrations. The focal length for the combined system isgiven by:

1

f=

1

fo+

1

ft− dot

foft(2)

where f is the focal length of the combined system, fo is thefocal length of the zoom lens, ft is the focal length of thefocus-tunable lens and dot is the distance between the prin-cipal planes of the zoom lens and the focus-tunable lens.The value of ft(t) is a time-varying function as control cur-

Focal Dist. = 1m

t1t2

t3

t4

t1 t2 t3 t4Temporal Sequences

1m / 30° / 15° 1m / -30° / 0° 2m / -30° / 0° 3m / 0° / 0°

Spatial Positions

Focal Dist. = 3m

Dist. = 2mFocal

Figure 4. Spatial and temporal multiplexing imaging for iriscapture. Four pair of eyes at different spatial positions are se-quentially captured by joint control of the liquid lens and steeringmirror. From the capture starting time ti, it may take several trialsto acquire one qualified image before refocusing to the next sub-ject. The first and second pairs of eyes are both located at 1mdistance with different view angles. All four pairs of eyes are cap-tured using the same image sensor in a controlled sequence.

rent changes. When the object distance is 5000mm, theextended back focal length is derived as 247mm consider-ing fo = 350mm. When the object distance is much largerthan objective lens, the focal length of the tunable lens ft(t)is independent of object distance o.

To evaluate the impact of the liquid lens on image res-olution, we have compared the iris image with and with-out it at 3m and 5m two distances as shown in Figure 5.Although less photons are collected after the beam passesthrough the extra focus-tunable lens resulting in a relativelydark iris image. It is demonstrated that the details of theiris structure are preserved regardless of standoff distanceand curvature of the liquid lens membrane. Even the fo-cal stacks are not simultaneous captured, it is feasible touse this focus-tunable lens to capture multiples eyes in real-time while maintaining the high resolution quality of eachimage due to its fast response.

3.2. 2D Steering Mirror for FoV Extension

Besides the fast focus-tunable method for DoF extensionalong depth direction, it is also important to expand hor-izontal and vertical views for iris imaging of unrestrictedstanding position. Based on the fundamental reflection law,we have introduced a 2D light beam steering method forFoV expansion by multiplexing imaging into one sensor ina temporal sequence. Such a fixed inverted lens architec-ture has been widely used in microscopy imaging and laserbeam control. Compared with the traditional PTZ methodto rotate the heavy zoom lens, the 2D motorized high reflec-

tive mirror offers precise and fast manipulation of light raydirection. Due to the small diameter, the common MEMSscanner and galvo mirror are excluded even they are fast upto several kilohertz. Considering our telephoto lens with anentrance pupil up to 73mm for reflection light collection,we have designed a large size mirror driven by two highspeed motors with exceptionally large tilt angle.

The new 2D steering device will steer the reflected lightfrom human iris in the addressed imaging volume back intothe sensor chip on focus. This customized 2D steering mir-ror can scan large 2D angles which is literally ±180◦ inhorizontal and over ±60◦ in vertical. Compared with thesmall 3.8◦x3.8◦ view angle of telephoto zoom lens at 5m,this new design is able to capture a participant standingat any position as long as the infrared lighting covers. Tomaximize the entrancing light, we have designed the reflec-tion mirror as 70mmx60mmx10mm size with high re-flectance (≥ 95%) gold coating for 850nm infrared light.The angular resolution of horizontal and vertical is 0.01◦

with a rotational speed up to 3500 rpm. The control sig-nal pulse frequency is up to 1MHz which delivers fast re-sponse capability.

Combining a 2D steering mirror with an electricallyfocus-tunable lens, our all-in-focus capture volume ex-tended iris imaging system enables precise and fast lightbeam collection from any area within the addressable 3Dspace. Taking advantage of the extremely fast system re-sponse, the participants at various spatial positions are se-quentially captured to form up the real-time frame sequenceas shown in Figure 4. It is noted that the four participantslocated at different distance (1-3m) and different angles (-30◦ to 30◦). By jointly adjusting the focus-tunable lens and2D steer device, our imaging system sequentially captureall four eyes from t1 to t4. From the imaging staring timeof one participant ti, it may take several frames to capturea qualified image before adjusting the device for next par-ticipant because of user cooperation and scene change. Thetargeted iris image will come into focus at least one frameof the temporal focal stacks. The spatial resolution of ourvolume extension device is majorly determined by the stepresolution in focus-tunable lens and angular resolution ofthe 2D steering mirror. The temporal resolution is deter-mined by the time response of the focus-tunable lens andthe steering mirror as well as the image sensor exposuretime. When it comes to a moving object, this imaging sys-tem will become effective only if equipped with fast sceneunderstanding method and precise control algorithm.

4. EXPERIMENTAL RESULTSIn this section, we first verify that our proposed all-in-

focus camera can be used for iris imaging DoF extensionwith a great capture volume. The DoF measurements oforiginal telephoto lens and focus-tunable lens mounted af-

(a) 3 m without liquid lens

(c) 5 m without liquid lens

(b) 3 m with liquid lens

(d) 5 m with liquid lens

Figure 5. Comparison of iris image quality captured at two dis-tances to evaluate the effect of focus-tunable lens. The iris im-ages (b) and (d) acquired using liquid lens appear darker due toinfrared light transmission loss.

1 1.5 2 2.5 3 3.5 4 4.5 5Distance (m)

0

500

1000

1500

2000

2500

3000

3500

4000

Dep

th o

f Fie

ld (m

m)

Original DoFExtended DoF

FDoF (mm)

RDoF (mm) 200 202 692 838 1500 1888 2390 2699 2700

300 771 935 1000 1142 1092 1111 1200605

DoF (mm) 500 1463 1773 2500 3030 3482 3810 3900807

Distance (m)

Dep

th o

f Fie

ld (m

m)

Figure 6. Evaluation of DoF extension capability by mountingthe Optotune focus-tunable lens between telephoto zoom lensand image sensor. The original telephoto zoom lens almost has afocal length less than 100mm across the range. The total extendedDoF is the sum of front DoF (FDoF) and rear DoF (RDoF). TheDoF can be extended up to 3.9m when the focus distance is 5m.The extended rear DoF is larger than the front DoF as expected.Each calculated DoF value shown in this image is the average offive repeated tests.

ter telephoto lens are compared across 1-5m range. Follow-ing this analysis, we evaluate the Hamming distance of pro-cessed iris image during DoF extension at three distances.The spatiotemporal multiplexing imaging is demonstratedfor auto-focusing among multiple participants within theextended capture volume. We conclude this section by de-scribing the potential of our 3D light beam steering systemin continuous iris imaging of a moving participant.

4.1. DoF Extension Measurement

To evaluate the DoF extension capability, we conduct anexperiment to calculate the actual DoF after focus-tunable

-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8Defocus (m)

0

0.2

0.4

0.6

0.8

1

Ham

min

g D

ista

nce

(a)

Distance from Focal Plane (m)

Ham

min

g D

ista

nce

-1 -0.5 0 0.5 1 1.5 2 2.5 3Defocus (m)

0

0.2

0.4

0.6

0.8

1

Ham

min

g D

ista

nce

(b)

Distance from Focal Plane (m)

Ham

min

g D

ista

nce

-1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3Defocus (m)

0

0.2

0.4

0.6

0.8

1

Ham

min

g D

ista

nce

(c)

Distance from Focal Plane (m)

Ham

min

g D

ista

nce

5 m

3 m

1 m

0.7 m

1 m

1.6 m

2.0 m

3 m

4.5 m

3.8 m

5 m

7.7 m

Figure 7. Hamming distance evaluation at various defocus dis-tances during DoF extension. Three DoF extension tests are con-ducted at different focus distances where (a) focal plane at 1m, (b)focal plane at 3m, and (c) focal plane at 5m. The image with anHD value below the dash line HD = 0.32 is considered as a matchto the template image captured at the original focal plane. Both. The extended DoFs are approximately 0.9m at 1m distance,2.5m at 3m distance and 3.9m at 5m distance respectively whenwe apply HD for evaluation. All images are evaluated without anyenhancement processing.

lens is mounted compared with the original DoF of the tele-photo lens. The image quality is measured according tothe ISO standard in which the lateral resolution across theiris diameter should be at least 200 pixels. (In most casesthe iris region resolution ≥150 pixels would be acceptable.Here we use a higher threshold for good recognition perfor-mance.) Every single image also has to pass the iris image

quality test in which a self-developed algorithm in accor-dance with ISO standard is applied. The participant wasasked to stand at 9 testing positions from 1 to 5m with a0.5m interval to repeat the test 5 times. Starting from onedesignated position, we will control the focal plane of thetelephoto zoom lens and the focus-tunable lens as the par-ticipant moves back and forth. Iris image quality check isconducted after all testing images are saved during post-processing.

It is noted that the 200 pixel requirement mainly re-stricts the rear DoF extension because the iris image resolu-tion decreases when liquid lens control current is adjustedfrom 0mA to negative values. And the image quality re-quirement majorly regulates the front DoF extension. Be-cause when we adjust the liquid lens control current from0mA towards larger positive value, the astigmatism is get-ting worse even the iris image is still focusable leading toblurred and distorted capture. The maximum DoF exten-sion appears when the largest focal length 350mm is usedto capture the target at 5m distance as shown in Figure 6.The extended DoF value 3900mm (front DoF = 1200mand rear DoF = 2700mm) is 37.5 times of the original DoF104mm from the telephoto lens. It takes about 80ms tocontrol the liquid lens focal plane moving through the wholeextended DoF range until the higher order oscillations fullysettled. If applying a low-pass filtered step signal, the set-tling time can be further reduced by 50% which is close tothe time-consuming for 1 frame in real-time capture.

4.2. Analysis of Hamming Distance

Since an acceptable iris images has to be recognizable,the image quality evaluation of iris capture after DoF doesnot represent the actual extension capability from the per-spective of recognition algorithm. In this study, we use thenormalized Hamming distance (HD) which determines thesimilarity of two iris templates as the metric to evaluate theactual recognizable DoF range [7]. Following the generaliris recognition framework, we preprocess the acquired irisimages including segmentation, normalization, and featureextraction using our self-developed methods. After iris fea-tures are quantized into binary codes, the normalized HD isintroduced as a quantitative measurement tool by bit-wiseoperation between two iris codes. Smaller HD values meanthat the two iris image are from the same class with a greaterprobability. And if the HD value is higher indicating a lowerlikelihood of the two images belong to the same class. Ac-cording to Daugman’s extensive investigation [6], the HDvalue must be below 0.32 to statistically minimize the pos-sibility of a false match between an enrolled iris templateand a test iris template.

Here we conduct an experiment to analyze DoF exten-sion capability at 1m, 3m and 5m. The iris image cap-tured at the best focal position as treated as the iris template

(a)

#1

#2

(b)#1

Subject #1：

Subject #2：

Height = 1540 mmDistance = 4380 mm

Height = 1800 mmDistance = 6430 mm

(c)#2

Figure 8. Demonstration of iris auto-refocusing between twoparticipants for height and distance adaption. High quality irisimages are captured for two participants at two positions with dif-ferent heights by jointly adjusting the focus-tunable lens and the2D steering mirror (a). Subject # 1 is sitting at 4380mm with aheight of 1540mm (b) and subject # 2 is standing at 6340mmwith a height of 1800mm (c). Image is cropped for better demon-stration which is not in accordance with the dimension scale.

enrollment and the iris images captured at the defocus po-sitions (±0.1m interval) are the test templates to comparewith the enrollment. Figure 7 shows the HD values whenthe test subject is standing at various defocus positions. Itis noted that every HD value is calculated from the averageof 5 repeated tests. The red point with a HD value belowthe 0.32 threshold means that iris template matches the en-rolled iris template. Even with careful control, the imagequality variation of each test does exist due to lighting, facepose and body movement. As expected, a larger DoF is de-rived compared with the DoF measurement based on imagequality standard.

4.3. Multi-Person Iris Auto-Refocusing

By combining the focus-tunable lens and 2D steeringmirror together, it is possible to capture the iris images ofmultiple participants at different locations with one camera.The fast response of our system enables the possibility ofreal-time multi-person iris imaging. We captured a sceneof two people locating at different distance (4380mm and6430mm) from the imaging system with different heights(1540mm and 1800mm) as shown in Figure 8. As long asthe distance and angle information fed into the control sys-tem, the steering mirror and liquid lens are simultaneouslyadjusted to direct the focused light rays of the target eyeson to imaging plane. Our system overcomes the obstaclessuch as height limit and position restriction in traditionallong-range iris imaging system. It is experimentally foundthat iris image sequence capture is necessary in order to ac-

quire at least one qualified testing sample. Besides the timeconsumption in hardware control, the efficiency of prepro-cessing methods including depth measurement, person de-tection and image quality evaluation also plays a major rolein system performance. Further improvement could also fo-cus on optimization of capture sequence strategy based ondynamic scene understanding.

4.4. Applications in Iris Recognition

In the same duration of time, larger capture volume al-lows more information collection of participants which in-crease the throughput of a biometric recognition system.This is also the research motivation of the first Iris on theMove (IOM) system back to 2006 [14]. To make full useof the advantages of our 3D light beam steering iris imag-ing system, we also conduct an experiment to capture theiris image of a moving subject. We use the novel capturevolume expanded camera with adaptively integrated light-ing to replace the IOM design of lens with fixed focal dis-tance and infrared lighting door. While a participant walksat a constant speed (around 1m/s) within the permissibleDoF, our system actively captures a sequence of iris imageframes in a continuous manner. Figure 9 shows that 3 fo-cused frames are grabbed out of the 15 sequential framesduring the walking period. If we look into the frames rightbefore and after the focused frame t7 at 3m, both t6 andt8 frames are blurry due to the unpredictable head move-ment while walking. The pivotal issue here is the imagingcontrol algorithm for successful capture. The sensor expo-sure time needs to be an appropriated value. Long-exposurewill introduce motion blur in a dynamic scene and short-exposure will transport insufficient photons to image sensorchip. In addition, we used a combination of fast focal sweepand auto-focusing control methods to follow the head mo-tion trajectory of our target participant. The overview ofthis all-in-focus iris imaging system and demo videos in-cluding DoF extension, fast focal plane switch by manipu-lating focus-tunable lens, multi-person iris auto-refocusingrecognition, and iris recognition of a moving participant arereleased on our lab website.

Despite the relatively rudimentary control algorithms,our results demonstrate the future of novel optical designwith multiplexing imaging in iris recognition applications.Compared with the IOM system, our capture volume ex-tended system provides a much higher throughput value to-wards unconstrained environments. It is believed that 3Dlight beam steering imaging method has potential applica-tions in biometric recognition such as iris recognition onthe move, continuous recognition, multi-person iris recog-nition, and omni-directional iris recognition.

t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 t13 t14 t15

3.3m 2.5m3mt3 t6 t7 t8 t13Reject Reject

Δt = 33ms

Figure 9. Demonstration to capture iris on the move using our proposed imaging system. 15 sequential frames are acquired while theparticipant moves at a constant 1m/s speed towards the imaging system. Three qualified iris images are captured at 3.3m, 3m and 2.5mas shown. The time interval between two captures is ∆t = 33ms. The exposure time is set as 3ms to sharply catch such a dynamic scenewith proper illumination.

5. CONCLUSION

In this paper, we demonstrate an active iris capture sys-tem with the greatly extended DoF up to 3.9m and theFoV up to ±180◦(H) x ±60◦(V). We develop this proto-type all-in-focus iris camera by combining a two-axis beamsteering mirror and the focus-tunable lens integrated withthe inverted telephoto zoom lens. Our large aperture de-formable optics provides high resolution iris image across awide depth range without sacrificing photons transport. Thefast response time of this electrically tunable lens enablesreal time refocusing by controlled focal sweeps. To captureiris images at various spatial positions in a temporal multi-plexing way, we engineered a motorized 2D reflection mir-ror with precise and fast beam control which outperformstraditional iris imaging methods in FoV extension. Besidesthe great capture volume extension, our imaging systemachieves high spatial, angular and temporal resolutions be-cause of innovative hardware design. Experimental resultsshow that the extended DoF is a factor of 37.5 comparedwith a conventional long focal lens when focus distance is5m. And the Hamming distance analysis further confirmsa larger DoF range does not sacrifice the quality of iris im-age for recognition. Our proposed system can dramaticallyincrease the throughput of a biometric recognition systemsince it is verified in multi-person eye auto-refocusing andiris on the move applications.

Our research concludes that this iris imaging volume ex-tension camera has a great potential to inspire novel bio-metric applications such as continuous recognition, omni-directional recognition, and active recognition. We expectthat the performance of our iris imaging system can be fur-ther improved by introducing adaptive illumination unit andreplacing steering module with a MEMS mirror for agile

capture control. In the following study, we have the plan todevelop precise synchronous control method and efficientscene understanding algorithms such as object detection,depth measurement, pose estimation, and semantic segmen-tation. It is also necessary to collect sufficient iris data indifferent scenarios to verify the performance of our all-in-focus system statistically. The multi-person iris recognitionon the move and iris recognition in surveillance scene willbe other interesting topics to explore using spatiotemporalmultiplexing imaging method.

Acknowledgements

This work was supported by the National Key Researchand Development Project (Grant No.2017YFB0801900),Tianjin Key Research and Development Project (GrantNo.17YFCZZC00200) and Science and Technology Coop-eration Project with Academy and University of SichuanProvince (Grant No.18SYXHZ0015) .

References

[1] F. Bashir, P. Casaverde, D. Usher, and M. Friedman. Eagle-eyes: a system for iris recognition at a distance. In 2008IEEE Conference on Technologies for Homeland Security,pages 426–431, 2008.

[2] M. Blum et al. Optotune focus tunable lenses and laserspeckle reduction based on electroactive polymers. In Pro-ceedings of SPIE vol. 8252, MOEMS and Miniaturized Sys-tems XI, 2012.

[3] Y. Bok, H. Jeon, and I. S. Kweon. Geometric calibrationof micro-lens-based light field cameras using line features.IEEE transactions on pattern analysis and machine intelli-gence, 39(2):287–300, 2016.

[4] G. Bub, M. Tecza, M. Helmes, P. Lee, and P. Kohl. Tem-poral pixel multiplexing for simultaneous high-speed, high-resolution imaging. Nature Methods, 7:209–211, 2010.

[5] C. Chong, K. Isamoto, and H. Toshiyoshi. Optically modu-lated mems scanning endoscope. IEEE Photonics Technol-ogy Letters, 8(1):133–135, 2006.

[6] J. Daugman. The importance of being random: statis-tical principles of iris recognition. Pattern Recognition,36(2):279–291, 2003.

[7] J. Daugman. The Essential Guide to Image Processing Chap-ter 25 - How Iris Recognition Works. Academic Press, 2009.

[8] L. Dong et al. Extended-depth-of-field object detection withwavefront coding imaging system. Pattern Recognition Let-ters, 125:597–603, 2019.

[9] T. Funahasahi, M. Tominaga, T. Fujiwara, and H. Koshimizu.Hierarchical face tracking by using ptz camera. In SixthIEEE International Conference on Automatic Face and Ges-ture Recognition, pages 427–432, 2004.

[10] S. Hsieh, Y. Li, C. Tien, and C. Chang. Extending the cap-ture volume of an iris recognition system using wavefrontcoding and super-resolution. IEEE Transactions on Cyber-netics, 46(12):3342–3350, 2016.

[11] International Organization for Standardization. Informa-tion technology—biometric data interchange formats—part6: Iris image data. ISO/IEC 19794-6, 2005.

[12] M. F. Land and D. E. Nillson. Animal Eyes. Oxford Press,2nd edition, 2012.

[13] X. Lv, W. Liao, K. Yi, and J. Gao. Design of a long distancezoom lens for iris recognition. In Chinese Conference onBiometric Recognition (CCBR), pages 476–485, 2018.

[14] J. R. Matey et al. Iris on the Move: Acquisition of Imagesfor Iris Recognition in Less Constrained Environments. Pro-ceedings of the IEEE, 94(11):1936–1947, 2006.

[15] J. R. Matey and L. R. Kennell. Iris recognition–beyond onemeter. In Handbook of Remote Biometrics, pages 23–59,2009.

[16] D. Miau, O. Cossairt, and S. K. Nayar. Focal sweep videog-raphy with deformable optics. In IEEE International Con-ference on Computational Photography (ICCP), pages 1–8,2013.

[17] K. Nguyen, C. Fookes, R. Jillela, S. Sridharan, and A. Ross.Long range iris recognition: A survey. Pattern Recognition,72:123–143, 2017.

[18] NIST. Iris Exchange (IREX) II - Iris Quality Calibration andEvaluation (IQCE). Technical report, National Institute ofStandards and Technology, 2011.

[19] R. J. Plemmons et al. Computational imaging systems foriris recognition. Proceedings of the SPIE, Advanced SignalProcessing Algorithms, Architectures, and ImplementationsXIV, 5559, 2004.

[20] K. N. Smith, V. P. Pauca, A. Ross, T. Torgersen, and M. C.King. Extended evaluation of simulated wavefront codingtechnology in iris recognition. In 2007 First IEEE Interna-tional Conference on Biometrics: Theory, Applications, andSystems, pages 1–7, 2007.

[21] J. Stirman, I. Smith, M. Kudenov, and S. Smith. Widefield-of-view, multi-region, two-photon imaging of neuronal

activity in the mammalian brain. Nature Biotechnology,34:857–862, 2016.

[22] S. Venugopalan, U. Prasad, K. Harun, K. Neblett,D. Toomey, J. Heyman, and M. Savvides. Long range irisacquisition system for stationary and mobile subjects. In In-ternational Joint Conference on Biometrics (IJCB), pages 1–8, 2011.

[23] B. Wilburn et al. High performance imaging using large cam-era arrays. In ACM SIGGRAPH, pages 765–776, 2005.

[24] S. Yoon, H. G. Jung, J. K. Suhr, and J. Kim. Non-intrusiveiris image capturing system using light stripe projection andpan-tilt-zoom camera. In IEEE Conference on Computer Vi-sion and Pattern Recognition, pages 1–7, 2007.

[25] C. Zhang, G. Hou, Z. Zhang, Z. Sun, and T. Tie. Efficientauto-refocusing for light field camera. Pattern Recognition,81:176–189, 2018.

[26] C. Zhou, D. Miau, and S. Nayar. Focal sweep camera forspacetime refocusing. Technical report, Department of Com-puter Science, Columbia University, MA, 2012.

1 . Introduction2 . RELATED WORK3 . PROPOSED IRIS IMAGING SYSTEM3.1 . Focus-Tunable Lens for DoF Extension3.2 . 2D Steering Mirror for FoV Extension

4 . EXPERIMENTAL RESULTS4.1 . DoF Extension Measurement4.2 . Analysis of Hamming Distance4.3 . Multi-Person Iris Auto-Refocusing4.4 . Applications in Iris Recognition

5 . CONCLUSION