Gabor Wavelet Based Vessel Segmentation in Retinal Images

M. Usman Akram∗, Anam Tariq§, Sarwat Nasir† and Shoab A. Khan‡

Computer Engineering∗, Software Engineering§,Computer Sciences and Engineering†, Electrical Engineering‡

College of Electrical and Mechanical Engineering∗,‡

National University of Sciences & Technology∗,‡

Fatima Jinnah Women University§, Bahria University†, Pakistan.Email: usmakram@gmail.com∗, anam.tariq86@gmail.com§, sarwatnasir83@gmail.com†, shoab@carepvtltd.com‡

Abstract— Retinal image vessel segmentation and theirbranching pattern are used for automated screening and diag-nosis of diabetic retinopathy. Vascular pattern is normally notvisible in retinal images. We present a method that uses 2-DGabor wavelet and sharpening filter to enhance and sharpen thevascular pattern respectively. Our technique extracts the vesselsfrom sharpened retinal image using edge detection algorithmand applies morphological operation for their refinement. Thistechnique is tested on publicly available DRIVE database ofmanually labeled images. The validation of our retinal imagevessel segmentation technique is supported by experimentalresults.

I. INTRODUCTION

D IABETES affects almost every one out of ten people,and has associated complications such as vision loss,

heart failure and stroke [1]. Diabetic eye disease refers to agroup of eye problems that people with diabetes may faceas a complication. Patients with diabetes are more likely todevelop eye problems such as cataracts and glaucoma, butthe disease’s affect on the retina is the main threat to vision[2]. Complication of diabetes, causing abnormalities in theretina and in the worst case blindness or severe vision loss,is called Diabetic Retinopathy [2].

Diabetic retinopathy is the result of microvascular retinalchanges. In some people with diabetic retinopathy, bloodvessels may swell and leak fluid. In other, new abnormalblood vessels grow on the surface of the retina [3]. The retinais the light-sensitive tissue at the back of the eye. A healthyretina is necessary for good vision [4].

Retinal vascular pattern facilitates the physicians for thepurposes of diagnosing eye diseases, patient screening, andclinical study [4]. Inspection of blood vessels provides theinformation regarding pathological changes caused by oc-ular diseases including diabetes, hypertension, stroke andarteriosclerosis [5]. The hand mapping of retinal vasculatureis a time consuming process that entails training and skill.Automated segmentation provides consistency and reducesthe time required by a physician or a skilled technician formanual labeling [2].

Retinal vascular pattern is used for automatic generationof retinal maps for the treatment of age-related maculardegeneration [6], extraction of characteristic points of the

retinal vasculature for temporal or multimodal image regis-tration [7], retinal image mosaic synthesis, identification ofthe optic disc position [8], and localization of the fovea [9].The challenges faced in automated vessel detection includewide range of vessel widths, low contrast with respect withbackground and appearance of variety of structures in theimage including the optic disc, the retinal boundary and otherpathologies [10].

Methods based on vessel tracking to obtain the vascula-ture structure, along with vessel diameters and branchingpoints have been proposed by [11]-[16]. Tracking consistsof following vessel center lines guided by local information.In [22], ridge detection was used to form line elementsand partition the image into patches belonging to each lineelement. Pixel features were then generated based on thisrepresentation. Many features were presented and a featureselection scheme is used to select those which provide thebest class separability. Papers [17]-[20] used deformablemodels for vessels segmentation . Chuadhuri et al. [21]proposed a technique using matched filters to emphasizeblood vessels. An improved region based threshold probingof the matched filter response technique was used by Hooveret al. [23].

In this paper, we present the colored retinal image vesselsegmentation technique that enhances and sharpens the vas-cular pattern using 2-D Gabor wavelet and sharpening filters.Our techniques creates a binary mask for vessel segmentationapplying edge detection algorithm on sharpened retinal imageand a fine segmentation masks is obtained by applyingmorphological dilation operation.

The paper is organized in four sections. In Section II, aschematic overview of our implementation methodology isillustrated. Section II also presents the step by step techniquesrequired for colored retinal image vessels segmentation.Experimental results of the tests on the images of the DRIVEdatabase and their analysis are given in Section III followedby conclusion in Section IV.

II. BLOOD VESSEL SEGMENTATION

The monochromatic RGB retinal image is taken as aninput and 2-D Gabor wavelet is used to enhance the vas-cular pattern especially the thin and less visible vessels

978-1-4244-2760-4/09/$25.00 ©2009 IEEE

Authorized licensed use limited to: Bahria University. Downloaded on August 24, 2009 at 00:21 from IEEE Xplore. Restrictions apply.

Fig. 1. Complete flow diagram for vessels segmentation

are enhanced using Gabor wavelet [24]. Before extractingvessels from enhanced retinal image, the blood vessels aresharpened using sharpening filter [25]. Vessels segmentationbinary mask is created by detecting vessels edges fromsharpened image. The blood vessels are marked by themasking procedure which assign one to all those pixels whichbelong to blood vessels and zero to non vessels pixels. Finalrefined vessel segmentation mask is created by applyingmorphological dilation operator [25].

Figure 1 shows the complete flow diagram of proposedblood vessel segmentation technique.

A. Gabor Wavelet

We have used 2-D Gabor wavelet to enhance the vascularpattern and thin vessels [24]. 2-D Gabor wavelet is used dueto its directional selectiveness capability of detecting orientedfeatures and fine tuning to specific frequencies [24], [26].

• The continuous wavelet transform Tψ(b, θ, a) is definedin terms of the scalar product of f with the transformedwavelet ψb, θ, a using equation 1 [27]Tψ(b, θ, a) = C

−1/2ψ 〈ψb, θ, a|f〉

= C−1/2ψ a−1

∫ψ∗(a−1r−θ(x− b))f(x)d2x (1)

where f ∈ L2 is an image represented as a squareintegrable (i.e., finite energy) function defined over R2

and ψ ∈ L2 be the analyzing wavelet. Cψ , ψ, b, θ and adenote the normalizing constant, analyzing wavelet, thedisplacement vector, the rotation angle, and the dilationparameter respectively.

• It is easy to implemented wavelet transform usingthe fast Fourier transform algorithm. Fourier wavelettransform is defined using equation 2 [26].

Tψ(b, θ, a) = C−1/2ψ a

∫exp (jkb)ψ̂∗(ar−θk)f̂(k)d2k

(2)where j =

√−1, and the hat (ψ̂∗ and f̂ ) denotes a

Fourier transform.• The 2-D Gabor wavelet is defined as [27]

ψG(x) = exp(jk0x) exp(−12|Ax|2) (3)

where k0 is a vector that defines the frequency of thecomplex exponential and A = diag[ε−1/2, 1], ε ≥ 1 isa 2 × 2 diagonal matrix that defines the elongation offilter in any desired direction.

• For each pixel position and considered scale value, theGabor wavelet transform Mψ(b, a) is computed usingequation 4 [26], for θ spanning from 0o up to 170o atsteps of 10o and the maximum is taken.

Authorized licensed use limited to: Bahria University. Downloaded on August 24, 2009 at 00:21 from IEEE Xplore. Restrictions apply.

Mψ(b, a) = maxθ|Tψ(b, θ, a)| (4)

Figure 2(b) shows the enhanced vascular pattern afterapplying Gabor wavelet.

B. Sharpening Filter

We used unsharp masking [25] on enhanced vessel retinalimage to sharpen the vascular pattern. The application ofGabor wavelet on colored retinal image enhances the vascularpattern but the resulting image is a little blurred so wehave used unsharp filter to sharpen the vascular edges. Thishelps in reliable extraction of vessels from the colored retinalimage.

Figure 2(b) shows enhanced retinal image and 2(c) showssharpened retinal image. It is clearly visible that the vesselsare much more prominent than they were in the originalimage.

Fig. 2. (a) Original retinal image; (b) Enhanced retinal image; (c) Sharpenedretinal image

C. Vessel Segmentation Mask

Vessels segmentation mask is created by extracting vesselsboundaries using edge detection algorithm and then byapplying morphological dilation operator [25]. In this paper,we have used canny edge detector [28]. Canny operatorsimultaneously optimizes three criteria: detection criterion,localization criterion, and elimination of multiple responsesand these rules also become a standard to evaluate edgedetection algorithm performance [28].

The vessels extracted using edge detection algorithm arerefined using standard morphological dilation operator [25].Dilation will fill the gaps between vessel boundaries detectedby edge detection algorithm.

Figure 3(b) shows the hollow vessels extracted using edgedetection algorithm and 3(c) shows the result of applyingmorphological dilation on 3(b).

Fig. 3. (a) Sharpened retinal image; (b) Extracted vessels; (c) Morpholog-ical refined vessels

TABLE IVESSEL SEGMENTATION RESULTS (DRIVE IMAGES)

Average StandardSegmentation Methods Accuracy Deviation

2nd. Observer 0.9473 0.0048

Staal et al. 0.9441 0.0079

Soares et al. 0.9466 0.0055

Proposed Method 0.9469 0.0053

III. EXPERIMENTAL RESULTS

The tests of proposed technique are performed with respectto the vessel segmentation accuracy using publicly availableDRIVE database [29]. The DRIVE database consists of 40RGB color images of the retina. The images are of size768×584 pixels, eight bits per color channel. The image setis divided into a test and training set and each one contains 20images. The test set is used for measurement of performanceof the vessel segmentation algorithms. There are two hand-labeling available for the 20 images of test set made by twodifferent human observers. The manually segmented imagesby 1st human observer are used as ground truth and thesegmentations of set B are tested against set A, servingas a human observer reference for performance comparisontruth. The segmentations of set B are tested against those ofA, serving as a human observer reference for performancecomparison [27]. We compared the accuracy of proposedtechnique with the accuracies of the methods of Staal et al.[22] and Soares et al. [27]. Table I summarizes the results ofvessel segmentation for above mentioned methods. It showsthe results in term of average accuracy and their standarddeviation. Figure 4 illustrates the blood vessel segmentationresults for proposed method.

Fig. 4. Row 1: Original retinal images from DRIVE database, Row 2:Blood vessel segmentation mask for each retinal image

IV. CONCLUSION

In this paper, a 2-D Gabor wavelet based colored retinalimage blood vessel segmentation technique is proposed. Theproblem with retinal images is that the visibility of vascular

Authorized licensed use limited to: Bahria University. Downloaded on August 24, 2009 at 00:21 from IEEE Xplore. Restrictions apply.

pattern is usually not good. So, it is necessary to enhancethe vascular pattern. In this paper, vessels are enhancedand sharpened prior to their detection. Vessel segmentationmask is created and refined using edge detection algorithmand morphological dilation operator respectively. We havetested our technique on publicly available DRIVE databaseof manually labeled images. Experimental results show thatour method performs well in enhancing and segmenting thevascular pattern.

REFERENCES

[1] K.M. West, L.J. Erdreich and J.A. Stober , “A detailed study of riskfactors for retinopathy and nephropathy in diabetes,” Diabetes , vol. 29,pp. 501-508, 1980.

[2] E. J. Susman, W. J. Tsiaras, and K. A. Soper, “Diagnosis of diabeticeye disease,” JAMA, vol. 247, pp. 3231-3134, 1982.

[3] C. Sinthanayothin, J. F. Boyce, H. L. Cook, and T. H.Williamson,“Automated localization of the optic disc, fovea and retinal bloodvessels from digital color fundus images”, Br. J. Opthalmol., vol. 83,pp. 231-238, August 1999.

[4] J. KansKy, Clinical Opthalmology, Butterworh-Heinmann , London,1994.

[5] T. Teng, M. Lefley and D. Claremont, “Progress towards automateddiabetic ocular screening: A review of image analysis and intelligentsystems for diabetic retinopathy,” Med. Biol. Eng. Comput., vol. 40, pp.2-13, 2002.

[6] A. Pinz, S. Bernogger, P. Datlinger and A. Kruger, “Mapping the humanretina,” IEEE Trans. Med. Imag., vol. 17, no. 4, pp. 606-619, Aug. 1998.

[7] C.L. Tsai, C. V. Stewart, H. L. Tanenbaum and B. Roysam, “Model-based method for improving the accuracy and repeatability of estimatingvascular bifurcations and crossovers from retinal fundus images,” IEEETrans. Inf. Technol. Biomed., vol. 8, no. 2, pp. 122-130, Jun. 2004.

[8] M. Foracchia, E. Grisam, and A. Ruggeri, “Detection of the optic discin retinal images by means of a geometrical model of vessel structure,”IEEE Trans. Med. Imag., vol. 23, no. 10, pp. 1189-1195, Oct. 2004.

[9] H. Li and O. Chutatape, “Automated feature extraction in color retinalimages by a model based Approach,” IEEE Trans. Biomed. Eng., vol.51, no. 2, pp. 246-254, Feb. 2004.

[10] A. M. Mendona, A. J. Campilho, “Segmentation of Retinal BloodVessels by Combining the Detection of Centerlines and MorphologicalReconstruction.” IEEE Trans. Med. Imag., vol. 25, no.9, pp.1200-1213,Sep. 2006

[11] I. Liu and Y. Sun, “Recursive tracking of vascular networks inangiograms based on the detection-deletion scheme,” IEEE Trans. Med.Imag., vol. 12, no. 2, pp. 334-341, Jun. 1993.

[12] L. Zhou, M. S. Rzeszotarski, L. J. Singerman, and J. M. Chokreff, “Thedetection and quantification of retinopathy using digital angiograms,”IEEE Trans. Med. Imag., vol. 13, no. 4, pp. 619-626, Dec. 1994.

[13] O. Chutatape, L. Zheng, and S. M. Krishnan, “Retinal blood vesseldetection and tracking by matched Gaussian and Kalman filters,” inProc. 20th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBS98), vol.20, pp. 3144-3149, 1998.

[14] Y. A. Tolias and S. M. Panas, “A fuzzy vessel tracking algorithm forretinal images based on fuzzy clustering,” IEEE Trans. Med. Imag., vol.17, no. 2, pp. 263-273, Apr. 1998.

[15] A. Can, H. Shen, J. N. Turner, H. L. Tanenbaum, and B. Roysam,“Rapid automated tracing and feature extraction from retinal fundusimages using direct exploratory algorithms,” IEEE Trans. Inf. Technol.Biomed., vol. 3, no. 2, pp. 125-138, Jun. 1999.

[16] M. Lalonde, L. Gagnon, and M.-C. Boucher, “Non-recursive pairedtracking for vessel extraction from retinal images,” Vision Interface,pp. 61-68, 2000.

[17] T. McInerney and D. Terzopoulos, “T-snakes: Topology adaptivesnakes,” Med. Image Anal., vol. 4, pp. 73-91, 2000.

[18] R. Toledo, X. Orriols, X. Binefa, P. Radeva, J. Vitri and J. Villanueva,“Tracking of elongated structures using statistical snakes,” in IEEEComput. Soc. Conf. Comput. Vision Pattern Recog. (CVPR), 2000.

[19] A. Vasilevskiy and K. Siddiqi, “Flux maximizing geometric flows,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 24, no. 12, pp. 1565-1578,Dec. 2002.

[20] D. Nain, A. Yezzi, and G. Turk, “Vessel segmentation using a shapedriven flow,” in Med. Image Comput. Comput.-Assist. InterventionMICCAI, 2004, pp. 51-59.

[21] S. Chaudhuri, S. Chatterjee, N. Katz, M. Nelson, and M. Goldbaum,“Detection of blood vessels in retinal images using two-dimensionalmatched filters,” IEEE Trans. Med. Imag., pp. 263-269, 1989.

[22] J. Staal, M. D. Abramoff, M. Niemeijer, M. A. Viergever and B.van Ginneken,“Ridge-based vessel segmentation in color images of theretina,” IEEE Trans. Med. Imag., vol. 23, no. 4, pp. 501-509, Apr. 2004.

[23] A. Hoover, V. Kouznetsova and M. Goldbaum, “Locating blood vesselsin retinal images by piecewise threshold probing of a matched filterresponse,” IEEE Trans. Med. Imag., vol. 19, no. 3, pp. 203-211, Mar.2000.

[24] A. Arnodo, N. Decoster and S. G. Roux, “A wavelet-based methodfor multifractal image analysis. I. Methodology and test applicationson isotropic and anisotropic random rough surfaces,” Eur. Phys. J. A,vol. 15, pp. 567-600, 2000.

[25] R. C. Gonzalez and R. E. Woods, Digital image processing, Prenticehall, second edition, 2002.

[26] J.P. Antoine, P. Carette, R. Murenzi, and B. Piette, “Image analysiswith two-dimensional continuous wavelet transform,” Signal Process.,vol. 31, pp. 241-272, 1993.

[27] J. V. B. Soares, J. J. G. Leandro, R. M. Cesar, H. F. Jelinek and M.J. Cree, “Retinal vessel segmentation using the 2-D gabor wavelet andsupervised classification.” IEEE Trans. on Med. Imag., vol. 25, no. 9,pp. 1214-1222, September 2006.

[28] R. C. Gonzalez, R. E. Woods and S. L. Eddins, Digital imageprocessing using matlab, Prentice hall, second edition, 2007.

[29] Niemeijer and B. van Ginneken, 2002 [Online]. Available:http://www.isi.uu.nl/Reseach/Databases/DRIVE/results.php

Authorized licensed use limited to: Bahria University. Downloaded on August 24, 2009 at 00:21 from IEEE Xplore. Restrictions apply.