Fringe noise removal of retinal fundus images using trimming regions - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 50

2 Methodology
An important yet overlooked stage in majority of the literature on image processing is the im-
age preprocessing step. Depending on the requirements of the study, and the main objectives
of the project, certain necessary steps and procedures should be performed in the image pre-
processing step so that the image is well prepared to fit the objective of further stages.
One of the procedures that should be implemented at this stage is the noise minimization.
A common problem in such cases is the noise which appears as illuminated bright spots on
the edges of the retinal images [ 25 ] . The bright fringes may have been formed due to the re-
fraction of the light within the eye, ambient light which might have affected the image if eye is
not placed properly in front of the device and also the eye response itself. An example of such
retinal responses can be viewed in Figure 2 .
FIGURE 2 Comparison of two retinal images, no fringe noise is present in left-hand image
while in the right-hand image the bright spot is completely visible.
The removal of fringe noise plays an important role in the accuracy of detection in particular
when the optic disk is the area of interest. This is due to the fact that the bright illuminated
fringe may inaccurately be located instead of the optic disk which is commonly known to be
as the brightest region in the retinal image.
There have been several approaches in order to enhance the detection precision in such in-
stances. Majority of the previous literature suggest the use of filters or adjust the contrast of
the image. Although the outcomes have been promising, many of the other crucial informa-
tion may have also been lost as the result.
In this study, the trimming regions have been used as the common approach. The trimming
region may be used to remove the unwanted noise and enhance the overall accuracy of detec-
tion. In order to remove all the noise, it should be noted that the radius of the proposed region
should be smaller than those of the actual region of interest so that all the fringe noises are
removed.
This approach was initially suggested by Zhang et al. [ 25 ] . They introduced a trimming
boundary represented by the following equations:
(1)