Fringe noise removal of retinal fundus images using trimming regions - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 49

processes have to ensure that the essential information needed for the following stages are not
destroyed or affected as the result. To achieve this, the RGB images are usually decomposed
to the primary colors and gray scaled. With the reduction in amount of data processed, the
overall processing time is also reduced. Similarly, other processes may also be implemented at
this stage including the implementation of different filters and adjustment of image contrast.
Once the images have been prepared, the feature localization step would be conducted in
order to localize the key features of interest. Accurate detection of the optic disk has been the
key feature of interest in this study. The noiseless solution for early prognosis of diseases such
as Glaucoma is presented further in this chapter.
This is then followed by feature extraction stage, where the information is extracted from
the regions of interest. At this stage, the overall objective of the study is fulilled.
Lastly, the detailed information obtained in the feature extraction step is studied in depth
and interpreted accordingly. The overall outcome and the key findings can also be displayed.
1.2 Retinal Image Processing
In this study, image-processing implementation on retinal images is considered. Majority of
the studies performed previously only consider or concentrate on the localizing, extracting
stages, and detecting features such as retinal blood vessels [ 8 - 12 ] or optic disk [ 13 - 17 ] . The
studies tend to ignore the preliminary stages of image processing, which could affect the over-
all accuracy and precision of the process.
Moreover, in automated systems, the input data may vary depending on the capturing in-
strumentation, their effect, response of the individual patients, image resolution, size and con-
trast which in turn could alter the inal results if the seting is chosen properly for localizing
and extracting stage.
In order to enhance the processing results and have a complete automated process, it is
essential to prepare the input images to the best possible format in the preprocessing stage.
There might also be times where a secondary image preprocessing stage is needed. This stage
could be implemented if the preliminary results do not produce the desired outcome.
These points highlight the fact that preprocessing stage is the basis and a crucial step in im-
age processing. As a result, for achieving the best possible outcomes a thorough and exact pro-
cedure should be suggested and conducted at this stage.
1.2.1 Ophthalmological Data
In order to start the image processing in ophthalmology, retinal images are needed. These im-
ages are captured using specialized instruments. However, for the purpose of this research
the open-source databases have been used. There are several different open-source databases
available online, including the digital retinal images for vessel extraction (DRIVE) [ 18 ] , struc-
tured analysis of the retina (STARE) [ 19 , 20 ], retinal vessel image set for estimation of width
[ 21 ] , retinopathy online challenge [ 22 ] , collection of multispectral images of the fundus [ 23 ] ,
and MESSIDOR database [ 24 ] .
For the purpose of this study, the DRIVE and STARE database has been chosen for further
analysis. The collection of DRIVE database was initiated by Staal et al. It contains 20 colored
retinal images, captured by Canon CR5 nonmydriatic 3CCD camera with a 45° field of view.
The images are digitized to 768 × 584 pixels, 8 bits per color channel [ 18 ] . STARE database con-
sists of about 400 retinal images and was initiated in 1975 by Michael Goldbaum. The images
are 8 bits per color plane at 605 × 700 pixels and were captured by TopCon TRV-50 fundus
camera with 35° field of view [ 19 , 20 ] .