Frequency domain analysis of grey level intensities for extraction of optic disc in retinal images

摘要

Revealing and extraction of optic disc (OD) in fundus images is the most important step in automatic screening system of diabetic retinopathy. The algorithm presented in this paper is focused on revealing and extraction of optic disc (OD) in fundus images. This algorithm adopts frequency domain approach to focus the OD and mathematical morphology for extraction of the OD. Large pathological signs, bright lesions limit the OD segmentation performance as brightness of both OD and bright lesions is similar. Bright lesions are extracted before localisation of the OD using DFT and morphological dilation is used for segmentation of the OD. This method was assessed on standard databases namely: DRIVE, DIARETDB1 and DIARETDB0. This algorithm acceptably recognises the OD in 255 out of 259 retinal images (98.45%) in 0.75 seconds.78.9-86.65% and 99.37-99.58% are the OD segmentation sensitivity and specificity achieved on these three databases. The proposed method demonstrates acceptable robustness on normal and pathological signs retinal images. Focused experimental results reveal the superior performance of presented work with respect to the methods demonstrated in literature.