Automated Segmentation Of The Optic Disc From Retinal Image Using Modified Local Intensity Clustering Model

Accurate optic disc (OD) segmentation plays an essential role in retinal disease diagnosis. In the paper, a novel method for segmenting OD boundary without manpower named automatic double boundary extraction is designed. It has two unique advantage. First, considering evolved contour always using imprecise initial contour which is manual or fixed, we propose a robust adaptive method for initializing the level set applying unsupervised machine learning theory. Second, in order to overcome complex OD appearance caused by some anomalies, the modified Local Intensity Clustering (MLIC) method by combining the multi-feature was proposed. Experimental results testing on publicly available DIARETDBO database demonstrate that our approach outperforms well known approaches.