Detection of blood vessels in retinal images based on top-hat transform / Syurga Fathonah Aris @ Azis

Retina is a part of the eye. Retina is the only location where blood vessels can be directly captured. Glaucoma and diabetic retinopathy are among the eye diseases which are related to retina, and serve as the main cause of blindness. Since most of the problems leading to the loss of vision are related to retina, it is important for an ophthalmologist to be able to do an accurate diagnosis so that any abnormality in retina can be detected earlier. Even though there are many different techniques have been proposed in order to detect the blood vessel in retinal image, there is still some issue on how accurate and fast the method works. Therefore, this project is done to be able to detect a blood vessel in retinal images by improving the previous method called a local entropy thresholding. The system is tested by using the sample from DRIVE and STARE database and is run by using MATLAB R2011b. The first step is done by extracting the green channel of RBG image since it gives better contrast for the retinal images. Next, the retinal image is enhanced by morphological transformation called opening by reconstruction. This basically removes small objects thus, clearing the image. Next, Top-Hat transform is applied to the retinal image. This transformation is useful for uncovering the details which are rendered invisible shading or illumination over the image. The next step is done by a mask generation and matched filter method. The process is then followed by applying a local entropy thresholding and length filtering for blood vessel extraction. The performance of the result is determined by calculating a sensitivity and accuracy which is compared with existing hand-labelled results from DRIVE and STARE database. The average ratio of accuracy and sensitivity for the proposed method is 0.89 and 0.71.