Modeling of the pixel based segmentation to detect Nerve Optic Head on the retinal image

Pixel based segmentation to detect the Nerve Optic Head (NOH) Pixels in the retinal image is proposed. Five main stages are required in the proposed model. They are image enhancement, binary thresholding, removing non-object pixels, finding Region of Interest, and dilation with mathematical morphology. Image enhancement stage is used to reduce the noise pixels and sharpened the target object. The enhanced image is transformed into a binary image in the second stage. Foreground pixels are then clustered or labeled using connected component, and the clustered pixels with fewer pixels are then removed. The density of the remained clustered pixels is then calculated to find the wide of the density are. The widest density is chosen as the ROI of NOH pixels. The last stage of the proposed model is dilation to enlarge the size of the ROI pixels. The best sensitivity, specificity, and balanced accuracy are 69.19%, 98.24%, and 83.72 % respectively. This accuracy is achieved by the mean filter in the enhancement stage.