Automatic Extraction and Localisation of Optic Disc in Colour Fundus Images

The color fundus images are used to track eye diseases by the ophthalmologists and it provides early signs of certain diseases such as diabetes. The optic disc extraction and segmentation is essential for the detection of many retinal diseases. The automatic screening facility helps the ophthalmologist to save time for the symptom detection. This paper presents a new histogram based multilevel thresholding method to segment and extract optic disc from colour fundus images. In this method, a hill-clustering technique is applied to the vessel free image histogram in order to determine the peak locations of the histogram and the histogram segments between the peaks are fitted by real rational functions. Finally, the multi level threshold values of the histogram are defined as the global minimum values of each rational function. This is an efficient technique for the segmentation because it is applicable to the value of image histogram directly to recognize the absolute transition point. The method was tested on a total of 404 images from publically available databases DRION, DIARETDB0 and DIARETDB1 and, also images collected from an ophthalmologist. The success rate of this method is 96.78%. The performance of the method was evaluated using the difference in centroid of the obtained optic disc and that of the ground truth obtained from an expert ophthalmologist. From the scatter plot, it is shown that the ground truth and detected optic disc centers have a high positive correlation.