Classification of Splat and GLCM Features in Fundus Images for Retinal Hemorrhage Detection

Image mining is the process of searching and discovering valuable information, knowledge in large volume of data. It is applied in Data mining, Image processing, Artificial Intelligence. Diabetic Retinopathy (DR) is eye disorder that affect retina, at severe stage it become vision loss. Early detection of DR is helpful to improve the screening system to prevent vision loss. Retinal hemorrhage is one kind of major abnormality to find the NonProliferative Diabetic Retinopathy (NPDR). The objective of our proposed work is to detect retinal hemorrhage for automatic screening of DR using Support Vector Machine (SVM) classifier. To detect retinal hemorrhage, retinal fundus images are taken from Messidor dataset. After pre-processing, retinal images using pixel of same color and intensity, the image is partitioned into non-overlapping area that covers the entire image. Splat and GLCM feature are extracted to improve the classification accuracy. In order to classify the given input images, different classes must be represented using relevant and significant features with the help of selection method that is processed by filter and wrapper approaches. Then hemorrhage affected retina is detected by SVM classifier. Finally classification accuracy is compared with K-Nearest Neighbor (KNN) classifier. Keywords— Non-Proliferative Diabetic Retinopathy (NPDR), Fundus image, Retinal Hemorrhage, Splat feature, Gray Level Co-Occurrences Matrix (GLCM).