Exudate Localization in Retinal Fundus Images Using Modified Speeded Up Robust Features Algorithm

摘要

Diabetic retinopathy (DR) is one of the retinal diseases caused due to blood leakage into the retina's tissues, such as artery blockage, blood leaks, and proteins in water. Many of the signs result in retina degradation, resulting in diminished sight or vision loss. Exudate (EX) is one of the early signs of DR. It is important to detect EX in the early stages to prevent vision loss. In this paper, a novel technique is proposed to localize the EX based on the modified speeded up robust features(SURF) method. The proposed method is organized in five steps, starting with the color space conversion, morphological operations, optic disc(OD) detection, SURF feature extraction, and EX localization. The proposed method is tested on both publicly available datasets and real-time datasets collected from Silchar Medical College and Hospital(SMCH). The proposed method achieved a pixel-level accuracy of 97%, which is better accurate than other state-of-the-art methods.