Robust optic disc detection based on multi-features and two-stage decision strategy

Abstract

This paper proposes a robust method based on multi-features and twostage decision strategy to locate the OD position. First, we use the global vessel distribution and directional characteristic and local appearance characteristic to find several OD candidates. Then we introduce the HOG to depict local details of OD candidates, and a SVM model is trained to classify OD and non OD candidate regions. Finally the correlation measure is used to remove redundant OD regions. This method has a good detection accuracy in normal images and diseased images. And the detection accuracy reached 97.9 % in four public image databases.