A Three Phases Procedure for Optic Disc Segmentation in Retinal Images

We introduce a new property based method for optic disc segmentation in retinal images. The suggested method involves parameters whose values can be easily modified to apply the method to retinal images with significantly different size. The method consists of three phases respectively dealing with image cropping, detection of light regions and optic disc segmentation. Cropping is done to reduce the computation burden. Two criteria dealing with the concentration of light pixels and with their distribution within the image are introduced during the second phase to detect a subset of the optic disc. The Circle Hough transform is used to perform segmentation, starting from the set identified during the previous phase. The performance of the method has been evaluated on public datasets. Comparisons with other methods in the literature in terms of average success, sensitivity, specificity, predictive value and overlap show that our method has comparable performance.