efficient iris segmentation based on eyelid detectionjestec.taylors.edu.my/vol 8 issue 4 august...

Journal of Engineering Science and Technology Vol. 8, No. 4 (2013) 399 - 405 © School of Engineering, Taylor’s University

399

EFFICIENT IRIS SEGMENTATION BASED ON EYELID DETECTION

ABDULJALIL RADMAN1,2,*, NASHARUDDIN ZAINAL

1, MAHAMOD ISMAIL

1

1Department of Electrical, Electronic & Systems Engineering, Faculty of Engineering & Built

Environment, Universiti Kebangsaan Malaysia, 43600, Bangi, Selangor DE, Malaysia 2Department of Communication and Computer Engineering, Faculty of

Engineering and Information Technology, Taiz University, Taiz, Yemen *Corresponding Author: [email protected]

Abstract

This paper proposes a computationally efficient eyelid detection algorithm for

detecting eyelid boundaries in iris images acquired under less constrained imaging conditions. The proposed eyelid detection algorithm is developed based on the

live-wire technique. The major advantage of the proposed algorithm is its

computational simplicity as compared to the prior eyelid detection algorithms.

The saturation color features of the sclera region of the HSI color space of the iris

image are exploited to determine the two intersection points between each eyelid

and the outer iris boundary. The strongly connected edges between these two points are detected using the live wire technique that is likely to be the eyelid

boundary. The experimental results obtained from UBIRIS.v1 database reveal that

the eyelid detection algorithm which proposed in this paper improves the

segmentation accuracy for the less constrained iris images.

Keywords: Iris segmentation, Eyelid detection, Live-wire.

1. Introduction

The rich and unique textural details of the iris make it the most promising

biometric trait for the personal identification [1]. The majority of iris recognition

systems use iris images taken under less constrained imaging conditions to

guarantee high performance segmentation. However, these constrained conditions

are not appropriate for security applications. Therefore, using unconstrained iris

images represents the promising solution for such kind of applications, but on the

other hand eyelid occlusion may degrade the performance of iris segmentation.

Removing eyelids occlusion in addition to localizing the inner and outer iris

400 A. Radman et al.

Journal of Engineering Science and Technology August 2013, Vol. 8(4)

boundaries represent the main stage in all iris recognition systems, which called

iris segmentation. Eyelid occlusion is one of the challenging noise factors in the

iris segmentation scenario. Most of the eyelid detection algorithms in the

literature are shape-based algorithms in which eyelid boundaries are detected as a

parabola or horizontal lines [2-5]. Active contour models [6, 7] are used to

localize iris boundaries, which make it possible to simultaneously localize the iris

and eyelid boundaries; but in contrast such methods require significant

computation time and many parameters must be carefully chosen to converge an

appropriate curve. In this paper, a shapeless edge-detection algorithm has been

proposed to localize eyelid boundaries. The optimal eyelid boundaries are

localized by means of the live-wire method [8]. The major advantage of the live-

wire method is its ability to be performed in real-time [9]. The proposed

algorithm outperforms on the existing eyelid detection algorithms in the following

advantages: it involves eyelid boundaries with irregular shapes, simple in

understanding and implementation, and significantly rapid.

The rest of this paper is organized as follows. In Section 2, the previous works

are presented. The eyelid detection algorithm using the live wire is described in

Section 3. The experiments and results, the discussion, and the comparison results

are presented in Section 4. Conclusions and future work are provided in Section 5.

2. Previous Works

Daugman [2] has improved his integro-differential operator [10] in order to localize

eyelid boundaries as a parabola; the circular path of the contour integration in the

original integro-differential operator has been changed to parabolic path. Wildes

[11] extracts the upper and lower eyelid boundaries with parabolic arcs also; the

parabolic Hough transform has been used to find eyelid boundaries on the edge map

of the iris image. Using the parabolic Hough transform [3, 4] localises eyelid

boundaries as a parabola. Masek [5] uses the linear Hough transform as a means of

localizing the upper and lower eyelids. The upper and lower eyelids are detected as

lines; then, horizontal lines are utilized to separate eyelid regions from iris region.

These horizontal lines intersect with first lines at the closest points to the pupil on

the outer iris boundary. Min and Park [12] utilizes the parabolic Hough transform as

a means of localizing eyelid boundaries in the normalized iris image instead of the

original iris image, in order to avoid the eye rotation problem and consequently

reduces the dimension of parameter space.

3. Eyelid Detection Algorithm using the Live Wire

This research work has proposed a computationally efficient eyelid detection

algorithm. The proposed eyelid detection algorithm is based on the live wire

technique [8]. Firstly, a radial edge detector based on the circle parameters of the

outer iris boundary is innovated to find the two intersection points between each

eyelid and the outer iris boundary. Through empirical analysis of the scalar

region, which contiguous the iris region, in the HSI color space; we found that the

scalar region is less saturated, which means that the saturation values of the scalar

region much closer to the zero. Thereby, the radial edge detector is used to scan

the outer iris boundary and compute the saturation average value of the ten

adjacent pixels for each pixel on the outer iris boundary; once the average value

Efficient Iris Segmentation based on Eyelid Detection 401


exceeds the threshold value at a certain pixel, this pixel is considered as an

intersection point between the eyelid and the outer iris boundary.

Figure 1 shows an example of detecting the four intersection points between

both the upper and lower eyelids and the outer iris boundary using the proposed

radial edge detector.

Fig. 1. An Example of Detecting the Four Intersection Points between

Eyelids and the Outer Iris Boundary using the Proposed Radial Edge Detector.

The eyelid detection algorithm in this paper is developed based on the live

wire method [8]; the live wire method requires two initial points to indicate the

start and end points of the eyelid. The two intersection points between each eyelid

and the outer iris boundary that have been detected by the radial edge detector are

used as initial points for the live-wire. In order to facilitate the detection process,

the two delimited regions which are used for eyelid boundary detection are

delineated based on the intersection points and the outer iris boundary as depicted

in Fig. 2. Towards delineate the optimal eyelid boundaries; the eyelid detection

region is first pre-processed to associate the edges with low costs. A combination

of edges of the Canny edge detector, gradient magnitude, gradient orientation, and

Laplacian zero-crossing are utilized to calculate these costs.

Fig. 2. The Delimited Regions to be used for Eyelid Boundary Detection.

The local cost function l (p, q) on the edge from pixel up to a neighboring

pixel q is a weighted sum of component cost functions as follows:

)(4.0)(1.0),(1.0)(4.0),( qfqfqpfqfqpl LMOC +++= (1)

where fC, fO, fM, and fL represent the Canny edge detector, gradient orientation,

gradient magnitude, and Laplacian zero-crossing cost functions, respectively. As

shown from Eq. (1), each cost term contributes by a constant weight to the total

cost function; the cost terms are designed to associate the edges that are likely to



be boundaries with low costs. More details on the live-wire operation and its’ cost

function construction can be found in [8, 13-15]. Automated detection of eyelids

boundaries using the proposed eyelid detection algorithm is illustrated in Fig. 3.

Fig. 3. Examples of Eyelid Boundary Detection using the Proposed Algorithm.

4. Experiments and Results

The proposed eyelid detection algorithm was tested on session 2 iris images of the

UBIRIS.v1 database [16], it includes 662 images. Several noise factors were

introduced, enabling the appearance of heterogeneous images regarding

reflections, contrast, luminosity, eyelid and eyelash iris obstruction and focus

characteristics [17]. We implemented the well-known iris segmentation method

proposed by Wildes [11] with the eyelid detection algorithm described in [11] and

the proposed eyelid detection algorithm. Next, the segmentation accuracy results

of both algorithms were compared to prove the effectiveness of the proposed

eyelid detection algorithm. The ground-truth images of the iris region for all iris

images in the session 2 of UBIRIS.v1 database were manually generated to

perform the evaluation process.

A quantitative evaluation of the classification error rate was adopted to

measure the level of iris segmentation accuracy. The classification error rate was

measured based on the comparison between the ground-truth image and the

binary output image which obtained by the iris segmentation algorithm. The

classification error rate (Ei) [18] measures the difference between the ground-truth

image (G(r,c)) and the binary output image (O(r,c)), which obtained by the iris

segmentation algorithm from the input image (Ii), as follows:

∑∑ ⊗×

== =

r

k

c

li lkGlkO

crE

1 1

),(),(1 (2)

where r and c refer to the image height and width, respectively. The comparison

between the output image and the ground-truth image was achieved by means of the

logical XOR operator, ⊗ . The overall classification error rate of the iris segmentation

algorithm (E) was calculated by the average errors of the input images:

∑==

N

iiE

NE

1

1 (3)

where N is the total of input images. Table 1 shows the segmentation accuracy

results obtained by the Wildes’ method [11] and the Wildes’ method with our

eyelid detection algorithm.



Table 1. Iris Segmentation Accuracy.

Method Segmentation

accuracy (%)

Wildes’ method 97.13

Wildes’ method with our

eyelid detection method 97.47

It can be observed from the segmentation accuracy summarised in Table. 1

that the proposed eyelid detection algorithm improves the segmentation accuracy

of the Wildes’ method.

Samples of iris segmentation results using the Wildes’ method with Wilde’s

eyelid detection algorithm and our eyelid detection algorithm are presented in

Fig. 4. As clearly observed from Fig. 4 the proposed eyelid detection algorithm

involves eyelid boundaries accurately regardless their shape.

(a)

(b)

Fig. 4. Samples of Iris Segmentation Results using the Wildes’ Method with

(a) Wildes’ Eyelid Detection Algorithm (b) our Eyelid Detection Algorithm.



5. Conclusions and Future Work

Eyelid occlusion represents one of the noise factors that degrade the performance of

iris segmentation. In this paper, we have proposed a simple and efficient eyelid

detection algorithm. The live-wire technique has been utilized to localize eyelid

boundaries based on the intersection points between the eyelid and outer iris

boundary. Results demonstrate that the proposed eyelid detection algorithm

improves the iris segmentation accuracy. In the future, the reflection noise can be

removed before detecting eyelid boundaries using the live-wire technique which

likely to be affected by such noise. This would improve the segmentation accuracy.

The proposed algorithm has designed to work with color images, but in the future,

our method will likely be used with single channel images.

Acknowledgement

This research has been conducted in the Computer & the Network Security

Laboratory, Universiti Kebangsaan Malaysia (UKM). The authors would like to

thank the University for sponsoring this research through research university grant

UKM-GUP-2011-060.

References

1. Radman, A., R.; Jumari, K.; and Zainal, N. (2011). Iris segmentation: A

review and research issues. Communication in computer and information

science. J.M. Zain et al. Ed., CCIS 179, 698-708.

2. Daugman, J.G. (2001). High confidence recognition of persons by iris

patterns. IEEE 35th International Carnahan Conference on Security

Technology, London, 254-263.

3. Jeong, D.S; Hwanga, J.W.; Kangb, B.J.; Parka, K.R.; Wonc, C.S.; Parkd, D.-

K.; and Kim, J. (2010). A new iris segmentation method for non-ideal iris

images. Image and Vision Computing, 28(2), 254-260.

4. Li, P.; Liu, X.; Xiao, L.; and Songet, Q. (2010). Robust and accurate iris

segmentation in very noisy iris images. Image and Vision Computing, 28(2),

246-253.

5. Masek, L. (2003). Recognition of human iris patterns for biometric identification.

B.Sc., The School of Computer Science and Software Engineering, The

University of Western Australia, Crawley WA, Perth, Australia.

6. Daugman, J. (2007). New methods in iris recognition. IEEE Transactions on

Systems, Man, and Cybernetics, Part B: Cybernetics, 37(5), 1167-1175.

7. Shah, S; and Ross, A. (2009). Iris segmentation using geodesic active

contours. IEEE Transactions on Information Forensics and Security, 4(4),

824-836.

8. Mortensen, E.; Morse, B.; Barrett, W.; and Udupa, J. (1992). Adaptive

boundary detection using `live-wire' two-dimensional dynamic programming.

Proceedings of Computers in Cardiology, 635-638.



9. Chodorowski, A.; Mattsson, U.; Langille, M; Hamarneh, G. (2005). Color

lesion boundary detection using live wire. Proceedings of SPIE 5747,

Medical Imaging 2005: Image Processing, 1589-1596.

10. Daugman, J.G. (1993). High confidence visual recognition of persons by a

test of statistical independence. IEEE Transactions on Pattern Analysis and

Machine Intelligence, 15(11), 1148-1161.

11. Wildes, R.P. (1997). Iris recognition: An emerging biometric technology.

Proceedings of the IEEE, 85(9), 1348-1363.

12. Min, T.-H.; and Park, R.-H. (2009). Eyelid and eyelash detection method in

the normalized iris image using the parabolic Hough model and Otsu’s

thresholding method. Pattern Recognition Letters, 30(12), 1138-1143.

13. Barrett, W.A.; and Mortensen, E.N. (1997). Interactive live-wire boundary

extraction. Medical Image Analysis, 1(4), 331-341.

14. Mortensen, E.N.; and Barrett, W.A. (1998). Interactive segmentation with

intelligent scissors. Graphical Models and Image Processing, 60(5), 349-384.

15. Radman, A.; Jumari, K.; and Zainal, N. (2013). Fast and reliable iris

segmentation algorithm. Image Processing, IET, 7(1), 42-49.

16. Proença, H.; and Alexandre, L.A. (2005). UBIRIS: A noisy iris image

database. 13th International Conference on Image Analysis and Processing,

3617 LNCS, Ed.: Springer Berlin / Heidelberg, 970-977.

17. Proença, H.; and Alexandre, L.A. (2006). Iris segmentation methodology for

non-cooperative recognition. IEE Proceedings- Vision, Image and Signal

Processing, 153(2), 199-205.

18. Proença, H.; and Alexandre, L.A. (2007). The NICE.I: Noisy Iris Challenge Evaluation - Part I. First IEEE International Conference on Biometrics:

Theory, Applications, and Systems, 1-4.

efficient iris segmentation based on eyelid detectionjestec.taylors.edu.my/vol 8 issue 4 august...

Documents