Feature Extraction in Digital Fundus Images

This paper proposes three novel approaches to extract the main features in color retinal images. The optic disk localization approach uses a method that finds the vessel branch with the most vessels. The proposed approach for optic disk boundary detection involves the use of color morphology in Lab space to have homogeneous optic disk region and geometric active contour model with variational formulation to extract the boundary of the optic disk. The exudate segmentation algorithm is composed of three steps. First, the image is preprocessed to enhance the contrast of the lesions. Next, the entropy feature is used to eliminate the optic disk. To extract the enhanced exudate regions, an approach based on spatially weighted fuzzy c-means clustering (FCM) is used. The spatially weighted FCM algorithm is formulated by incorporating the spatial neighborhood information into the standard FCM algorithm. The success rates of disk localization and disk boundary detection are 99.7% and 96%, respectively. The sensitivity and specificity of exudate detection are 100% and 89%, correspondingly. The detection and analysis could be applied to automatic mass screening and diagnosis of the retinal diseases.