Superpixel Classification Based Optic Cup Segmentation

In this paper, we propose a superpixel classification based optic cup segmentation for glaucoma detection. In the proposed method, each optic disc image is first over-segmented into superpixels. Then mean intensities, center surround statistics and the location features are extracted from each superpixel to classify it as cup or non-cup. The proposed method has been evaluated in one database of 650 images with manual optic cup boundaries marked by trained professionals and one database of 1676 images with diagnostic outcome. Experimental results show average overlapping error around 26.0% compared with manual cup region and area under curve of the receiver operating characteristic curve in glaucoma detection at 0.811 and 0.813 in the two databases, much better than other methods. The method could be used for glaucoma screening.