• Osareh, A.
• Mirmehdi, Majid
• Markham, R.
• Thomas, B.

Abstract

Aim:To provide an automatic method for the localisation of theoptic disk in retinal images.Method: The location of the optic disk is an important issue inretinal image analysis as it is a significant landmark feature andits diameter is usually used as a reference length for measuringdistances and sizes. Inmany cases, this process can bestraightforward and a circular Hough transform or edge detection andwatershed segmentation have been attempted. When there are otherfeatures in the retina images, such as Exudates (EXs),these mayexhibit similar spectral characteristics to the optic disk. We haveproposed a method to identify EXs automatically [1] that alsopartially extracts the optic disk as candidate EX regions due tocolour similarity between the EXs and optic disk. This method isbased on colour normalisation, contrast enhancement and coloursegmentation based on Fuzzy C-Means (FCM) clustering. This partiallocalisation of the optic disk requires further processing to isolateit. In this paper, we report on the selection of candidate optic diskregions amongst the EXs, boundary analysis, and optic disk centre andradius estimation using minimum boundary arc lengths.Results: We applied our proposed method to 50 colour retinal imagesand the optic disk was identified correctly in all the images. Ourmethod provides an accurate circular approximation of the optic diskregion suitable forapplications such as ours [1]. We intend to usea post-processing step for further, precise segmentation using""snakes"" in future, as our method provides an automatic bootstrapsnake spline for the process.Conclusion: In this work we introduce an efficient approach toaccurately localise the optic disk. This robust technique will beused as part of our aim to establish a cost effective mass screeningsystem for diagnosingdiabetic maculopathy.

Topics

• impedance spectroscopy
• clustering