Splat Feature Classification With Application to Retinal Hemorrhage Detection in Fundus Images

Abstract

In this paper, a splat feature classification method for retinal hemorrhage detection is presented. The detection of retinal hemorrhages is most important in the automated screening system to find out the diabetic retinopathy diseases. Here the retinal images are divided into non-overlapping segments. For each segment, contains pixels with same color and spatial location. A set of features is extracted to describe its characteristics such as splat area, splat extent, splat orientation, texture features etc. These features are selected by using the filter approach followed by Wrapper approach. After the feature selection step, the classifier is used to calibrate the feature space and its discriminative characteristics. The Accuracy of splat labels are finding out by the classification techniques and detect the retinal hemorrhages in a fundus images.