A Comparison of Local Descriptor-based Data Augmentation Techniques for Glaucoma Detection using Retinal Fundus Images

Glaucoma is an ocular disorder which may cause permanent blindness. Early diagnosis of glaucoma may aid in better treatment. In this paper, we develop an automated deep learning-based approach for diagnosis of Glaucoma using local descriptors-based augmentation techniques. More specifically, we use Local variance (LV) as data augmentation, along with AlexNet and we compare its performance with local binary pattern (LBP) based data augmentation for glaucoma detection. Our experiments are performed on publicly available RIM-ONE r2 dataset. Our results suggest that LBP-based data augmentation provides better classification accuracy than LV. Specifically, it achieved a classification accuracy of 96.70% with AlexNet and outperformed the existing approaches which performed experiments on same dataset.