TELKOMNIKA
Retinal Image Figure 1.
2.1.1 Background Estimation
A color retinal imag background. This dark backg
feature extraction and lesion d is important to remove the bac
We present a local me creates a binary background s
greater than threshold value, belongs to background.
The algorithm for the b
Step 1 : Divide the acquired re
Step 2 : Compute the local me
Step 3 : Use local mean value
stdI using equation 2
Step 4 : Select threshold value
Step 5 :
for each pixel,
Calculate stdI
T
if true , add pixe
if false
, add pixe
end_for
2.1.2 Noise Segmentation M Noise in color retinal
distorted. Both seem to exi illumination is usually adequa
located near the edge of the r abnormality detection. That is
abnormalities.
In our technique, w area and it is applied on reti
steps i.e. feature extraction convert RGB Red, Green, Blu
the way a human experience space [19]. The RGB retinal im
ISSN: 1693-6930
e Preprocessing: Background and Noise Segmenta 1. Flow Diagram of Retinal Image Preprocessing
ion
age consists of a semi circular region of int kground is initially never really black. It is not ne
n detection algorithms on this area and it consume ackground from input retinal image.
mean and variance based method [19] for backgro d segmentation mask by applying a threshold on std
e, the block is considered as original retinal image e background extraction mask is as below:
retinal image into non-overlapping blocks ean value
MI using equation 1
ue computed in step 2 to compute the local standar n 2
lue empirically Threshold?
ixel in original retinal image area pixels ixel in background area pixels
Mask
al image is normally due to noise pixels and pixe exist in regions where illumination has been in
quate in the center of the image, poor image qu e retinal image. Regions with poor image quality m
t is why they should be detected and removed be we create binary noise segmentation mask which i
etinal image to ensure not to process the noisy a n and abnormality detection. In this segmentatio
Blue retinal image into HSI color space because fi ces colors and secondly noise can be easily remo
l image is converted into HSI color space using equ
ntation Ibaa Jamal 539
interest on a dark necessary to apply
mes more time so it round estimation. It
stdI. If the stdI is ge area otherwise it
1 dard deviation value
2
ixels whose color is inadequate. Since
quality regions are may cause errors in
before detection of h includes the noisy
y area in upcoming tion technique, we
firstly it is closer to moved in HSI color
quations 3, 5 and 6
TELKOMNIKA
Vol. 10, No. 3 540
.
where
here R, G and B represent RE
The algorithm for noise remov
Step 1
: Divide the input retina
Step 2 : Use histogram equaliz
Step 3
: Use a 3x3 median filte
Step 4 : Convert the equalized
Step 5 : Calculate
N noise fac
Step 6 : Select a threshold valu
Step 7
:
for
each pixel, Calculate
NI Thres
if true
, add pixel in no
if false , add pixel in
end_for
Figure 2 shows the images. These segmentation
removed in fine segmentation.
Figure 2. Coarse segmentatio
2.2 Fine Segmentation