Fast optic disc segmentation in retinal images using polar transform

Abstract

Glaucoma is one among major causes of blindness. Early detection of glaucoma through automated retinal image analysis helps in preventing vision loss. Optic Disc segmentation from retinal images is considered as the preliminary step in developing the diagnostic tool for early Glaucoma detection. A novel hierarchical technique for optic disc localization and segmentation on retinal fundus images is presented in this paper. Retinal vasculature and pathologies are delineate and removed by using morphological operations as preprocessing steps. Circular Hough transform is used to localize the optic disc. Region of interest is calculated and a novel polar transform based adaptive thresholding is performed to obtain the precise boundary of optic disc. The methodology has shown considerable improvement over existing methods in terms of accuracy and processing time. The algorithm is evaluated on a number of publicly available retinal image sets which includes MESSIDOR, DIA-RETDB1, DRIONS-DB, HRF, DRIVE and RIM-ONE, with average spatial overlap approximately 85%.