Robust hemorrhage detection in diabetic retinopathy image

Abstract

To improve the robust performance to detect hemorrhage lesions in diabetic retinopathy image, a background estimation and vessel exclusion based algorithm is proposed in this paper. Candidate hemorrhages are located by background estimation and Mahalanobis distance, and then on the basis of shape analysis, vessel exclusion is conducted to remove non hemorrhage pixels. Experiments results show that the performance of our method is effective to reduce the false negative results arise from inaccurate vessel structure.