Fusing Geometric and Appearance-based Features for Glaucoma Diagnosis

In this paper, we propose to fuse geometric and appearance-based features at the feature-level for automatic glaucoma diagnosis. The cup-to-disc ratio and neuro-retinal rim width variation are extracted as the geometric features based on a coarseto-fine localization method. For the appearancebased feature extraction, the principal components analysis is adopted. Finally, these features are combined at the feature-level based on the random projection and the total error rate minimization classifier. Experimental results on an in-house data set shows that the feature-level fusion can enhance the classification performance comparing with that before fusion.