Automatic rapid optic disc detection from infrared radiation retinal fundus images

Optic disc (OD) detection is a main step in computer-aid eye disease diagnosis. This paper presents an integrated technique to detect OD from retinal fundus images automatically, accurately and rapidly. The method is especially robust to infrared radiation (IR) fundus images, which may suffer terrible imaging artifacts, such as extremely low contrast, overexposed edges, or invisibility of blood vessels. The detection system integrates down-sampling for acceleration, a novel equalization of histogram for contrast enhancement, a series of filters within polar coordinate system for circular brightness feature detection, an improved connected component extraction for post-processing and a size-adaptable filter for outputting. Tested over Clinical dataset, the proposed method outperforms the earlier methods by achieving a total accuracy of 93% (94% for IR subset) and an average time consuming of 0.59 second.