ISSN : 0975-2927
EISSN : 0975-9166
RAJAPUT G.G.1, RESHMI B.M.2*, SIDRAMAPPA C.3
1Department of Computer Science, Gulbarga University, Gulbarga, Karnataka, India
2Department of Computer Science, Gulbarga University, Gulbarga, Karnataka, India
3Department of Computer Science, Gulbarga University, Gulbarga, Karnataka, India
* Corresponding Author : breshmi@yahoo.com
Received : 06-11-2011 Accepted : 09-12-2011 Published : 12-12-2011
Volume : 3 Issue : 4 Pages : 172 - 179
Int J Mach Intell 3.4 (2011):172-179
DOI : http://dx.doi.org/10.9735/0975-2927.3.4.172-179
Conflict of Interest : None declared
Acknowledgements/Funding : The authors are grateful to the referees for their helpful comments. This work is supported by UGC, New Delhi under Major Research Project grant in Science and Technology (F.No. 40-257/2011 (SR) dated 29.06.2011). The authors are grateful to UGC for their
In this work, we present a fovea center localization method in digital color photographs of the retina (color eye fundus images). The proposed method is based on prior knowledge of optic disk center and optic disk diameter. The detection of this anatomical feature is a prerequisite for the computer aided diagnosis of several retinal diseases, such as Diabetic Maculopathy. The proposed method is evaluated on 33 retinal photographs from public DRIVE data set. The experimental results demonstrate that the proposed method is able to detect the fovea center by providing encouraging results.
Optic disk, Macula, Fovea, mathematical morphology, diabetic maculopathy.
The macula-center (fovea) is an important anatomical landmark in automated analysis of diabetic maculopathy in color fundus photographs [1] . The macula is the darker region of the retina. It is the area providing the clearest, most distinct vision. The center of the macula is called the fovea, an area where all of the photoreceptors are cones; there are no rods in the fovea. The fovea is the point of sharpest, most acute visual acuity. The very center of the fovea is the foveola. [Fig-1] illustrates a typical retinal image showing marked fovea region. In physical terms, the fovea region is a circle of 0.25mm of diameter, with its center located a number of optic disk diameters away (e.g. 2 disk diameters) from the optic disk center, in the temporal side of the optic nerve (i.e. towards the macula center). Because of its important function in vision, the distance at which lesions are located from the fovea influences their clinical relevance [2] . This work is part of a larger project to develop an automated screening system for diabetic retinopathy in general and diabetic maculopathy in particular. Diabetic retinopathy is a common complication of diabetes and the largest cause of blindness and vision loss in the working population of the western world [3] . Hard exudates appear as bright lesions in eye fundus images and are the most commonly found retinal abnormalities. However, the detection of hard exudates is not sufficient to detect and grade the Diabetic Maculopathy [4] , since the distribution of these exudates around the fovea also must be considered. [Fig-2] illustrates the polar coordinate system (which is centered on the fovea center) used to analyze the distribution of the exudates in the fovea.
Compared to the publications that dealt with location of the optic disk [5] , there are few publications for fovea detection. To detect macula and fovea, a template (a Gaussian blob) matching approach was used by Sinthanayothin et al. [6] . Narasimha-Iyer et al [7] used a two step approach, which is based on the optic disk diameter, a region of interest and an adaptive threshold. An appearance based method for fovea detection is presented by Singh et al. [8] . Niemeijer et al. [9] proposed a method based on cost function as well as a point distribution model to detect and locate the fovea. Li and Chutatape [10] combined the information provided by low intensity pixels (characteristics of the fovea region) and main blood vessels arcade, to detect the fovea with a parabola fitting method. Sekhar et al. [11] proposed the use of Hough transform and some morphological operations to automatically detect the fovea. Soumitra Samanta et al. [12] proposed a fovea detection algorithm based on the structure of the blood vessels, little bit information about optic disk and mathematical morphology. Bob Zhang et al. [13] discussed fovea detection method based on optic disk vessel candidate detection, and optic disk vessel candidate matching.
In this paper, we have localized the fovea center based on optic disk (OD) center, optic disk diameter and mathematical morphology. This work is part of a larger research project to design a method for grading diabetic maculopathy with respect to fovea centre. The proposed method localizes the fovea center as a single pixel in the fovea region (fovea centriod) that is used for grading diabetic maculopathy and is presented in two stages. In first part, the OD center and optic disk diameter is determined. In the second stage, using the information of OD center and its diameter the ROI for the macula region is located and then the fovea center is detected. [Fig-3] shows the diagram of the proposed methodology for detection fovea center.
This paper is organized as follows. The next section provides the information about materials used. Next the proposed method is described. Then the experiment results are discussed. Finally conclusion is given in last section.
In this work the digital color fundus photographs are selected from DRIVE dataset [14] . A total of 33 retinal images of dimensions 768 × 584, captured by a Canon CR5 non-mydriatic 3CCD camera with a 45 degree field of view (FOV) are used for evaluation of the proposed method.
The proposed method uses two stages, Optic disk detection and the fovea center detection. The following sections describe these two stages in detail.
Our proposed method needs optic disk center and its diameter. Detection of optic disk center is based on morphology. The optic disk is a brightest component of the fundus image and is clearly visible in the red channel. Histogram equalization is applied on the red channel of the image for contrast enhancement. Then the image is inverted. The regions with minimal intensities in the region using extended minima transform are identified. The extended-minima transform is the regional minima of h-minima transform [15] This transformation is a thresholding technique that brings most of the valleys to zero. The h-minima transform suppresses all the minima in the intensity image whose depth is less than or equal to a predefined threshold. The output image is a binary image with the white pixels represents the regional minima in the original image. Regional minima are connected pixels with the same intensity value, whose external boundary pixels all have a higher value. The output is a binary image. The extended minima transform on the f image with threshold value, h is shown in (1).
E = EM (f, h) (1)
Where E is the output image.
The selection of threshold is very important where the higher value of h will lower the number of regions and a lower value of h will raise the number of regions. In this work h value (threshold height) is selected empirically and is set to 20. The result is shown in [Fig-4] (e). Morphological opening is applied using disk shaped structuring element of size 8 to eliminate the regions that are wrongly located. The mean intensities of the identified regions are computed. The region with the lowest mean intensity is then selected as the optic disk region and centroid is computed. Knowing the OD center, the diameter of the OD is computed. The optic disk measures about 1.5 mm in diameter. In the image the maximum diameter of the optic disk can be 100 pixels [16] . [Fig-4] shows the approach used for optic disk center detection. The algorithm for optic disk detection is given below.
Input: Color Fundus Image
Output: Optic disk region
Step 1: Retrieve the red channel of the image.
Step 2: Enhance the contrast of the image by using histogram equalization and invert the image.
Step 3: Identify the regions with minimal intensities in the region using extended minima transform (Empirical threshold, h=20).
Step 4: Apply opening operation using disk shaped structuring element of size 8 to eliminate the regions those are wrongly located.
Step 5: The mean intensities of the identified regions are computed. Select the region with the lowest mean intensity as the optic disk region and plot the centroid. Using this centriod and Optic disk diameter 100 pixels, we draw the circle to locate the OD region.
After locating the optic disk center and knowing the diameter of the optic disk, the fovea center can be determined by setting an area of restriction in the vicinity of the image center, as determined by the optic disk center. The distance and position of macula with respect to the diameter of the optic disk remains relatively constant. Siddaligaswamy. et. al. [17] presented a method for detecting ROI for the macula. In our work the same approach was used to localize the ROI for macula before detecting fovea center. It is situated about 2 disk diameter temporal to the optic disk in fundus images and the mean angle between macula and the center of the optic disk against the horizon is -5.6 + or – 3.3 degrees. Since the location of macula region varies from individual to individual, a ROI for the macula is localized as shown in the [Fig-5] . The width of the ROI is taken equal to 2DD as the mean angle between the fovea and the center of the optic disk to the horizontal, as mentioned varies between -2.3 to -8.9. degrees. Macula is a darkest component of the fundus image and is clearly visible in green channel. Histogram equalization is applied on the green channel of the image for contrast enhancement. The ROI for the macula is localized as mentioned above. This is the search area to detect fovea center in the macula. The regions with minimal intensities in the search region using extended minima transform (empirical threshold height, h=50) as explained in the previous section are identified. Opening operation is applied using disk shaped structuring element of size 8 to eliminate the regions that are wrongly located. The mean intensities of the identified regions are computed. Then we select the region with the lowest mean intensity as the macula region and then the centriod is computed. [Fig-6] shows the approach used for fovea center detection. Thus the algorithm for detection of fovea center is given below.
Input: Color fundus image
Output: Fovea region
Step 1: Get the green channel of the image.
Step 2: Enhance the contrast of the image using histogram equalization.
Step 3: ROI for macula region is localized.
Step 4: Identify the regions with minimal intensities in the ROI using extended minima transform. (Empirical threshold, h=50).
Step 5: Opening operation is applied using disk shaped structuring element of size 8 to eliminate the regions those are wrongly located.
Step 6: The mean intensities of the identified regions are computed. Then we select the region with the lowest mean intensity as the fovea region and compute its centriod.
The proposed method has been evaluated on total of 33 images not affected by pathologies taken from DRIVE database. The fovea center was detected successfully in all the images of the data set with a success rate of 100%. Fig. (7) illustrates the method of detecting fovea centre for a sample image. Few images showing fovea candidate region and fovea center obtained by applying the proposed method are given in [Fig-8] and [Fig-9] respectively. The proposed algorithm for locating fovea center in the fundus eye image is simple and effective compared to the other methods in the literature [7-8] . Continuing the proposed work we are carrying out experiments on images from other data sets containing diabetic lesions.
A novel approach has been proposed to locate macula centre (fovea) automatically in color fundus retinal photographs based on prior knowledge of optic disk centre, its diameter, and mathematical morphology. The visible macula region provides better feature information in diagnosing diabetic maculopathy. The result shows that the proposed method is able to detect fovea centre on 33 retinal images of DRIVE data set correctly. Experiments are being carried out on other retinal photographs to study the detection results on larger set of images containing diabetic lesions.
The authors are grateful to the referees for their helpful comments. This work is supported by UGC, New Delhi under Major Research Project grant in Science and Technology (F.No. 40-257/2011 (SR) dated 29.06.2011). The authors are grateful to UGC for their financial support.
[1] Rajendra Acharya U., Eddie Y.K. Ng, Jasjit S. Suri., (2008) Arttech House.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[2] ETDRS report, (1991) 98, 766 – 785.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[3] Klonoff D., Schwartz D., (2000) Diabetes Care, 3(3) 390–404.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[4] Nayak J, Bhat P.S., Acharya U.R., (2009) Journal of Medical Engineering & Technology, 33(2) 119- 29.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[5] Abdel-Razik Youssif A.A.H., Ghalwash A.Z., Abdel-Rahman Ghoneim A.A.S. (2008) IEEE Transactions on Medical Imaging, 27(1) 11–18.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[6] Sinthanayothin C., Boyce J. and Williamson C.T., (1999) Br. J. Ophthalmol., 38, 902–910.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[7] Narasimha-Iyer H., Can A., Roysam B., Stewart C.V., Tanenbaum H.L., Majerovics A., Singh H. (2006) IEEE Transactions on Biomedical Engineering, 6, 1084-1098.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[8] Singh G.D., Joshi J., Sivaswamy., (2008) IEEE International Conference on Image Processing, IEEE, San Diego, CA, USA.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[9] Niemeijer M., Abràmoff M.D., Ginneken B.V. (2007) IEEE Transactions on Medical Imaging, 26, 116–127.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[10] Li H., Chutatape O. (2004), IEEE Transactions on Biomedical Engineering, 51, 246–254.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[11] Sekhar S., Al-Nuaimy W., Nandi A. (2008), 6th European Signal Processing Conference, Lausanne, Switzerland.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[12] Soumitra Samanta, Sanjoy Kumar Saha, (2011), Bhabatosh chanda, Second International Conference on Engineering Applications of Information Technology, 206-09.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[13] Bob Zhang, Fakhreddine Karray (2010), Proc. of First International Conference on Autonomous and Intelligent systems, Povoa de Varzim, Portugal, 1-5.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[14] http://www.isi.uu.nl/Research/Databases/DRIVE/
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[15] Soille P. (1999), Springer-Verlag, 170-171.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[16] Siddalingaswamy P.C., Gopalkrishna Prabhu K., (2010) International Journal of Computer Applications,. 1(7) 1-5.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
[17] Siddalingaswamy P.C., Gopalkrishna Prabhu K., (2010) Proc. of International Conference on Systems in Medical and Biology, India, 331-334.
» CrossRef » Google Scholar » PubMed » DOAJ » CAS » Scopus
 | Fig. 1- Retinal Image |
 | Fig. 2- Polar coordinate system of Fovea |
 | Fig. 3- Automatic Fovea center detection system |
 | Fig. 4- (a) Original image (b) Red channel (c) Contrast enhancement (d) Inverted image (e) After applying extended minima transform (f) After thresholding (g) Optic disk center marked (h) Optic disk region |
 | Fig. 5- Macula region search area |
 | Fig. 6- (a) Original image (b) Green channel (c) Contrast enhancement (d) OD center detection (e) Macula ROI (f) After applying extended minima operation (g) After thresholding (Fovea candidate region) (h) Fovea superimposed on original image (i) Fovea Center marked |
 | Fig. 7a- Macula ROI |
 | Fig. 7b- After applying Extended Minima operation |
 | Fig. 7c- Fovea on the original image |
 | Fig. 7d- After thresholding |
 | Fig. 7e- Fovea center marked |