Detection of pixel-based optical disc in retinal images

Diabetic Retinopathy (DR), a complication of diabetes, is one of the most common causes of blindness in adults. Early detection of DR disease is important to prevent visual loss. One of the main processing steps of early detection of DR is the detection of optical disc (OD) since it is similar to the lesion in retinal images in terms of brightness and it can be accepted as a noise that should be eliminated from the image. Therefore, the optical disc must be determined and then ignored from the image to be able to detect the lesions that are signs and symptoms of DR disease. In this study, in order to detect OD in retinal images, we proposed a novel filter which determines a reference point within optic disc by using morphological processes, detects 8 pointsat 45-degree angles between them around this reference point by moving the filter in the image and draws the circle passing through these points. For this aim, the green channel of the image in the RGB color space was converted to gray level and median filter was used to reduce the noise in the image. The contrast of the image was then adjusted and the image was improved by using the adaptive histogram equalization method. In pixel-based (OD) detection, the veins were removed from the image since they are noises for the target aims. Gabor and Top-hat transformations were used for the segmentation of the veins. Then, the improved images were converted to a binary level by using thresholding. A reference pixel (point) in the OD was determined by applying morphological processes into the binary image.8 linear pixel lines were created at 45-degree angles outward from the reference pixel. The difference between each pixel of these created lines and reference pixel was compared with pre-specified threshold value, when the difference exceeds to the threshold value the previous pixel was selected as a marker, then the OD was detected by drawing a circle passing through these marker pixels. As result, the implemented experiments show that our pixel-based filter designed in order to detect OD in retinal images reached to satisfactory results.