Automatic detection of glaucoma in retinal images

A new method to detect glaucoma is proposed in this paper, which is based on principle components analysis (PCA) and Bayes classifier. Firstly, optic disc center is located using the combination of thresholding and distance transformation. Eigenvector spaces of normal set and glaucoma set are obtained respectively using PCA. A test image is projected onto these two spaces and the distance between projection and each template is calculated. Finally, decision is made according to Bayes classifier. The success rate of optic disk localization is 95.3% and 89.9% for normal set and glaucoma set respectively. The glaucoma detection algorithm was tested by over three hundred retinal images and the success rate is 78%.