Hard Exudate Detection Using Local Texture Analysis And Gaussian Processes

Exudates are the most noticeable sign in the first stage of diabetic retinopathy. This disease causes about five percent of world blindness. Making use of retinal fundus images, exudates can be detected, which helps the early diagnosis of the pathology. In this work, a novel method for automatic hard exudate detection is presented. After an exhaustive pre-processing step, Local Binary Patterns Variance (LBPV) histograms are used to locally extract texture information. We then use Gaussian Processes to distinguish between healthy and pathological retinal patches. The proposed methodology is validated using the E-OPHTA exudates database. The experimental results demonstrate that Gaussian Process classifiers outperform the current state of the art classifiers for this problem.