International Conference on Digital Image Processing

Contrast Enhancement of Retinal Vasculature in Digital Fundus Image

*M. H. Ahmad Fadzil, **H. Adi Nugroho, H. Nugroho, I. Lila Iznita

Department of Electrical and Electronic Engineering
Universiti Teknologi PETRONAS
Bandar Seri Iskandar, 31750 Tronoh, Perak Darul Ridzuan, Malaysia
* fadzmo@petronas.com.my, ** hanungadin@gmail.com

Abstract—Analyzing retinal fundus image is important for
early detection of diseases related to the eye. However, in
fundus images the contrast between retinal blood vessels and
the background is very low. Hence, analyzing or visualizing
tiny blood vessels is difficult. Fluorescein angiogram overcomes
this imaging problem but it is an invasive procedure that leads
to other physiological problems. In this work, we enhance the
low contrast of retinal vasculature in the ocular fundus image
by determining the retinal pigments makeup, namely
haemoglobin, melanin and macular pigment using independent
component analysis. Independent component image due to
haemoglobin pigment obtained exhibits higher contrast retinal
vasculature. Even though the fundus fluorescein angiograph
method produces higher contrast images, results show that this
approach outperforms other noninvasive enhancement
method, such as contrast limited adaptive histogram
equalization and can be beneficial for vessel segmentation.
Contrast enhancement factor of 3.2 for digital retinal images is
achieved. This improvement in contrast reduces the need of
applying contrasting agent on patients.
Keywords-fundus images; retinal vasculature; contrast
enhancement; independent component analysis.
I. INTRODUCTION
Analyzing retinal fundus image is important for early
detection of several diseases related to eyes, such as diabetic
retinopathy. In diabetic retinopathy, retinal capillary
occlusion occurs and accordingly causes enlargement of
foveal avascular zone [1]. Foveal avascular zone is the fovea
where there is no blood vessels and located in the very centre
of macula. Digital images obtained from fundus camera are
shown in Figure 1. Figure 1(a) illustrates the problems of
very low contrast and non-uniform illumination which can
be seen at the area towards the edge of the image. Figure
1(b) shows the occurrence of noise which consists of impulse
and Gaussian noises. Detection of the foveal avascular zone
is even difficult due to very low image contrast of blood
vessels against the background in the macular area.
A number of enhancement methods focused in the image
spatial domain [2, 3]. Histogram equalization with its
modification is commonly used to enhance the image
contrast. However, histogram equalization tends to over-
enhance the image and results in noisy appearance of the
output image. One of the adaptive methods called contrast
limited adaptive histogram equalization (CLAHE) worked
well on the enhancement of retinal vasculature [4]. Iznita
found that the contrast improvement using contrast limited
adaptive histogram equalization on retinal image model
ranges between 1.7 and 3 [5]. However, contrast limited
adaptive histogram equalization creates artefacts in the
enhanced image and the selection of contrast gain limit is
image-dependent.
Figure 1. Digital retinal image obtained from fundus camera [6]
Colours observed in the retinal image correspond to the
architecture of retinal layer and the optical properties of the
pigments [7, 8]. Styles et al. developed a model using the
concentrations of the five main absorbers found in the fundus
layers, namely retinal haemoglobin, choroidal haemoglobin,
choroidal melanin, retinal pigment epithelium melanin and
macular pigment [9]. This approach focuses more towards
reconstructing rather than improving the contrast. Other
related works used the information of colour taken from
digital colour images to transform digital colour image
(RGB) into independent components that correspond to the
biological makeup of the skin [10, 11]. Tsumura et al.
showed that spatial distributions of melanin and
haemoglobin from a skin colour image can be separated
using independent component analysis [12]. Nugroho et al.
successfully applied a technique based on principal
component analysis and independent component analysis to
convert the RGB skin image into a skin image that represents
skin areas due to melanin and haemoglobin only [11]. The
above efforts focus on using independent component
analysis to transform digital colour image (RGB) into
independent components that correspond to the biological
makeup of the skin.
The objective of this work is to address the low contrast
problem of fundus images obtained from fundus camera
when no contrasting agent is injected. A novel approach is
presented to enhance the contrast by determining the retinal

978-0-7695-3565-4/09 $25.00 © 2009 IEEE 137

DOI 10.1109/ICDIP.2009.32
pigments, namely haemoglobin, melanin and macular
pigment from fundus images. Distribution of haemoglobin is
used to reveal retinal vasculature.
incident light reflected light
pupil plane
lens
ocular
media
inner limiting
membrane
retina
ocular retinal pigment
fundus epithelium
choroid
sclera
Figure 2. A model of ocular fundus showing pathways of reflected light.
II. APPROACH
The approach taken in this research is as follows. First, a
model of ocular fundus based on the light interaction is
developed to describe the reflectance of the fundus. Second,
a model of spectral absorbance of the retinal image is
developed to show the components composing the observed
colours in a digital fundus image. Third, independent
component analysis based on the spectral absorbance of the
model is applied to determine retinal pigments from fundus
images. Finally, a fundus image model is developed to test
performance of the proposed algorithm.
A. Ocular fundus model
Ocular fundus represents the structure of the back of the
eyes that consists of multiple layers of tissue [9]. An ocular
fundus image obtained by a fundus camera shows different
intensity of reflectance. The reflectance depends on the
wavelength, the structure of fundus’ layers, the optical
properties and quantities of retinal pigments in the ocular
fundus. The incident light from a fundus camera can be
reflected, absorbed, scattered or transmitted by the retinal
tissues.
Generally, the structure of the eye can be classified into
two main groups, namely ocular media and ocular fundus
[11]. Ocular media consists of cornea, lens and vitreous. It is
located between the ocular fundus and the observer. The
ocular fundus consists of retina, retinal pigment epithelium,
choroid and sclera. The reflectance of the fundus can be
described in the terms of these layers. Fig. 2 depicts a model
of ocular fundus showing possible pathways of the reflected
light.
B. Ocular fundus spectral absorbance model
Ocular fundus represents the structure of the back of the
eyes. The spectral absorbance image provides useful
138
information to identify the absorbance components. In this
work, we focus on the distribution of retinal pigments,
namely haemoglobin, melanin, and macular pigment, rather
than on the fundus layers, to model spectral absorbance of
the ocular fundus.
Basis of linear combination of the absorption coefficients
of melanin, haemoglobin and macular pigment is modeled
from three absorbances μa(λ1), μa(λ2) and μa(λ3) at three
wavelengths λ1, λ2 and λ3. These wavelengths λ1, λ2 and λ3
represent the red (R), green (G) and blue (B) colour
channels. Two conditions are assumed when analyzing
fundus spectral absorbances. First, the colour observed in the
fundus image is due to the distributions of melanin,
haemoglobin and macular pigment. Second, the quantities of
these components are spatially independent of each other.
The spectral absorbance in the fundus image represents the
linear combination of the absorption coefficients of melanin,
haemoglobin and macular pigment.
Let sx,y and vx,y designate a three-dimensional (3-D)
quantity vector and composite colour vector on an image
coordinate (x, y) of a digital colour image. A mixing matrix
A with a1, a2 and a3 represents pure colour vectors of the
three components (haemoglobin, melanin, macular pigment)
per unit quantity. It is assumed that a linear combination of
mutually independent pure colour vectors with the quantities
of s1x,y, s2x,y and s3x,y result in the composite colour vectors of
v1x,y, v2x,y and v3x,y on the image coordinate (x, y). The
following equation illustrates the transformation matrix,
where T denotes the transpose.
vx,y = A sx,y (1)
sx,y = [s1x,y, s2x,y, s3x,y]T (2)
The pixel value of each channel corresponds to each
element of the colour vector. Fig. 3 depicts the absorbances
of the ocular fundus which consist of pure spectral vectors of
melanin, haemoglobin, and macular pigment.
μa(λ3) Haemoglobin a1
Melanin a2
Distribution of fundus
s2x,y spectral absorbances
s1x,y
Macular
pigment a3
s3x,y
a4
μa(λ1)
μa(λ2)
Figure 3. Model of spectral absorbance of the ocular fundus
C. ICA of fundus spectral absorbance image
Independent component analysis (ICA) is a technique to
determine the original signals from mixtures of several
independent sources. The use of independent component
analysis is investigated to separate the spatial distributions of
melanin, haemoglobin, and macular pigment in the ocular
fundus. Three colour channels, namely red, green and blue,
are used to determine these components. By applying the
independent component analysis to the composite colour TABLE I. INTENSITY RANGE OF RETINAL VASCULATURE, MACULAR
vectors in the image, the relative quantity and pure colour REGION AND THE BACKGROUND IN RED, GREEN AND BLUE CHANNELS
vectors of each component are determined with no prior
information on the quantity as well as colour vector. The
Retinal vasculature Macular region Background
quantities of the pigments are presumed to be mutually Minimum R 102 114 175
independent for the image coordinate. The separating matrix Maximum R 255 212 255
W is defined to separate vector ŝx,y using the following Minimum G 15 53 76
equations. Maximum G 102 97 135
ŝx,y = W vx,y (3) Minimum B 21 22 28
ŝx,y = [ŝ1x,y, ŝ2x,y, ŝ3x,y]T (4) Maximum B 70 59 86
As illustrated in Fig. 3, the composite colour vector is
denoted as
vx,y = [μa(λ1), μa(λ2), μa(λ3)]T (5) Blood vessel model
Logarithm transformation is used to transfer reflectance
spectra into spectral absorbances [10].
[μa(λ1),μa(λ2),μa(λ3)]=[-log(rx,y),-log(gx,y),-log(bx,y)] (6)
Here, the values of rx,y, gx,y and bx,y correspond to pixel
value in the colour channels of red, green and blue,
Macular region
respectively. According to the model of spectral absorbance
in the ocular fundus from Fig. 3, the colour density vector of
the fundus can be stated as
vx,y = A sx,y + a4, (7) Background
where mixing matrix A represents pure colour vectors of
the three components (haemoglobin, melanin, macular
pigment) per unit quantity. It is assumed that a linear
Figure 4. Fundus image model
combination of mutually independent pure colour vectors
with the quantities of s1x,y, s2x,y and s3x,y result in the
composite colour vectors of v1x,y, v2x,y and v3x,y on the image III. RESULT AND DISCUSSION
coordinate (x, y). Additionally, a4 is similar to noise in the A fundus image model is firstly tested using independent
ICA model. In this case, the model is assumed to be noise- component analysis to see performance of the algorithm. The
free, therefore a4 can be eliminated. inputs to the FastICA are the red, green and blue channels
Several methods, such as fast fixed-point algorithm taken separately from the colour fundus image model. As
(FastICA) [13], joint approximate diagonalization of eigen- can be seen from Fig. 5, the proposed algorithm successfully
matrices (JADE) [14] and information-maximization separates the components into three, namely macular region,
(infomax) [15] have been proposed to solve the problem of retinal vasculature and the background. These three
independent component analysis. The FastICA algorithm is components represent the distribution of the macular
used to get the estimated independent components because pigment, the haemoglobin and the melanin.
of its good accuracy and high computational speed [13]. In this work, 25 retinal images containing macular region
D. Fundus image model are taken from DRIVE database [6]. Here a retinal image
showing macular region is shown undergoing the algorithm.
A fundus image is modeled to test the performance of The image is assumed to be noise-free. In a preliminary work
independent component analysis in separating the using the above algorithm, it is found that non-uniform
distribution of macular pigment, haemoglobin and melanin. illumination in fundus images resulted in false detection of
As shown in Table 1, the maximum and minimum intensity the retinal pigments. This is because the algorithm responds
range of retinal vasculature, macular region and the to the spectral reflectance or absorbance of the retinal
background in red, green and blue channels are taken from pigments in the image. Therefore, homomorphic filtering is
the 20 test images from DRIVE [6] to represent the performed prior to independent component analysis to
distribution of haemoglobin, macular pigment and melanin, reduce the problem of non-uniform illumination.
respectively. Homomorphic filtering is used to reduce illumination which
In Fig. 4, a fundus image model with uniform distribution varies slowly in space and at the same time, it enhances the
intensity of red, green and blue channels is shown and reflectance of the object which usually changes abruptly. A
consists of macular region, retinal vasculature and smaller region containing the macular area is also taken to
background. Using this model as an input, the independent see the enhancement of retinal capillaries, which usually
component analysis should be able to separate these show a very low contrast between retinal vasculature and the
components into three outputs, namely macular region, background.
retinal vasculature and background.

139
 a. Macular region b. Retinal vasculature
c. Background
Figure 5. Independent component analysis of a fundus image model
Fig. 6 shows the original image undergoing
homomorphic filtering and its independent components
estimated by the FastICA algorithm. The components
represent the distribution of the pigments, namely macular
pigment, haemoglobin and melanin. The brighter area in the
centre of the first independent component (Fig. 6(b))
represents the distribution of macular pigment. The second
independent component (Fig. 6(c)) shows the distribution of
haemoglobin. It is indicated by the enhancement of retinal
vasculature. The third independent component (Fig. 6(d))
shows brighter area related to the distribution of melanin.
This result is consistent with the location of melanin, which
is fairly distributed in the retinal pigment epithelium and the
choroid. Based on the assumption that the image is noise-
free, independent component analysis is able to determine
the retinal pigments.
As shown in Fig. 7, a green band image undergoing
contrast limited adaptive histogram equalization (CLAHE) is
compared to the haemoglobin-related component image after
the intensity is being inverted to demonstrate that contrast
enhancement is also achieved. Iznita reported that
enhancement factor as high as 3 can be achieved using
CLAHE [5]. In this work, CLAHE is also performed on the
same images undergoing the proposed algorithm to compare
the contrast improvement between these two methods.
Contrast improvement is measured as a ratio of contrast
value of the image undergoing the enhancement method and
that of the green band image. From the 25 images, the green
band image undergoing CLAHE and the proposed algorithm
shows the average contrast of 29.80 and 30, respectively.
Having measured the average contrast of the green band
image which is 9.50 as a reference, the proposed algorithm
with contrast enhancement factor of 3.2 shows a slightly
better improvement than that of the CLAHE with contrast
enhancement factor of 3.15. However, the proposed
algorithm produces no artefacts in the process.
140
a. Fundus image after b. First component
homomorphic filtering
c. Second component d. Third component
Figure 6. Independent component analysis of a retinal image
containing macular region
a. Green band image after CLAHE b. Independent component (haemoglobin)
Figure 7. Contrast enhancement of retinal vasculature
IV. CONCLUSION
Analyzing retinal fundus images is usually difficult as
they are of very low contrast. Low contrast between blood
vessels and the background makes it difficult to accurately
determine retinal vasculature. Retinal vasculature can be
used to determine existence of pathology, macular area and
foveal avascular zone. Typical contrast enhancement
methods usually create artefacts or introduce noise. Even
though fluorescein angiography produces better contrast
enhancement, it is not preferable due to its invasive nature of
injecting contrasting agent.
In this work, the developed method based on the spectral
absorbance model and independent component analysis
enables us to determine the retinal pigments, namely
haemoglobin, melanin and macular pigment. A fundus image
model has been developed to test the performance of the
proposed algorithm. As a result, retinal vasculature, macular
pigment and melanin distribution can be determined from
digital fundus image. The algorithm produces no artefacts in
the process. Using the distribution of haemoglobin, the
contrast between retinal vasculature and the background can
be enhanced. Contrast enhancement factor of 3.2 for digital
retinal images is achieved. This improvement in contrast [12] N. Tsumura, et al., “Independent-component analysis of skin color
reduces the need of applying contrasting agent on patients. image,” J. Opt. Soc. Am. A, vol. 16, no. 9, 1999, pp. 2169.
[13] A. Hyvarinen, “Fast and robust fixed-point algorithms for
independent component analysis,” Neural Networks, IEEE
REFERENCES
Transactions on, vol. 10, no. 3, 1999, pp. 626-634.
[14] J.F. Cardoso and A. Souloumiac, “Blind beamforming for non
[1] G.H. Bresnick, et al., “Abnormalities of the foveal avascular zone
Gaussian signals,” IEE Proceedings-F, vol. 140, no. 6, 1993, pp.
in diabetic retinopathy,” Arch Ophthalmol, vol. 102, no. 9, 1984,
362-370.
pp. 1286-1293.
[15] A.J. Bell and T.J. Sejnowski, “An information-maximisation
[2] M. Niemeijer, et al., “Comparative study of retinal vessel
approach to blind separation and blind deconvolution,” Neural
segmentation methods on a new publicly available database,”
Computation, vol. 7, no. 6, 1995, pp. 1129-1159.
Proc. Medical Imaging 2004: Image Processing, SPIE, 2004, pp.
pp. 648-656.
[3] T. Shimahara, et al., “Contrast enhancement of fundus images
using regional histograms for medical diagnosis,” Proc. SICE 2004
Annual Conference, 2004, pp. 650-653 vol. 651.
[4] M.H.A. Fadzil, et al., “Extraction and reconstruction of retinal
vasculature,” Journal of Medical Engineering Technology, vol. 31,
no. 6, 2007, pp. 435-442
[5] L. Iznita, “Analysis of Retinal Vasculature and Foveal Avascular
Zone for Grading of Diabetic Retinopathy,” Master thesis, M.Sc.
Thesis Electrical and Electronics Engineering Programme,
Universiti Teknologi Petronas, Bandar Seri Iskandar, 2006.
[6] J. Staal, et al., “Ridge-based vessel segmentation in color images
of the retina,” Medical Imaging, IEEE Transactions on, vol. 23,
no. 4, 2004, pp. 501-509.
[7] S.J. Preece and E. Claridge, Physics Based Medical Image
Understanding of the Colouration of the Ocular Fundus with
Application to Detection of Diabetic Retinopathy, Technical
Report CSR-04-08, Department of Computer Science, University
of Birmingham, 2002.
[8] F. Orihuela-Espina, et al., “Histological parametric maps of the
human ocular fundus: preliminary results,” Medical Image
Understanding and Analysis, 2003, pp. 133-136.
[9] I.B. Styles, et al., “Quantitative analysis of multi-spectral fundus
images,” Medical Image Analysis: Special Issue on Functional
Imaging and Modelling of the Heart (FIMH 2005), vol. 10, no. 4,
2006, pp. 578-597.
[10] N. Tsumura, et al., “Independent Component Analysis of Spectral
Absorbance Image in Human Skin,” Optical Review, vol. 7, no. 6,
2000, pp. 479-482.
[11] H. Nugroho, et al., “Determination of Skin Repigmentation
Progression,” Proc. Engineering in Medicine and Biology Society,
2007. EMBS 2007. 29th Annual International Conference of the
IEEE, 2007, pp. 3442.

141