Abstract
The Diabetic Retinopathy (DR) is a visual complication of diabetes and one of the principal cause of lost vision not recoverable in industrialized countries. It can be treated, if detected, in its first stages. This is not an easy task because the patients with DR do not perceive any symptoms until the visual loss develops in advanced stages when the treatment is less efficient. Hard exudates are the most common lesions in the early stages. In this work, we developed an automatic method for hard exudates detection in Diabetic Retinopathy images with an acceptable level of confidence that can help specialists in the diagnosis and screening of this disease. We also propose an exhaustive study of feature selection first by individual analysis and then by combining several features. This strategy is efficient, comparable and competitive with results of methods of the state of the art.