Feature Based Classification of Retinal Blood Vessels Using Multifractal Technique

In this work, multifractal method is used to segment and analyze human retinal fundus images. Both in normal and abnormal cases, the retinal images acquired under standard protocols undergoes segmentation process for extraction of retinal vasculature. From the segmented vessels, different performance measures are obtained by comparing the segmented image with ground truth images. Using Support Vector Machines (SVM), these significant performance measures along with the derived parameter of Vessel to Vessel Free (VVF) area ratio are further subjected to classification. By using this method normal and abnormal images can be differentiated. When compared to other kernels SVM classifier with order 3 polynomial kernel gives better performance. The proposed study seems to be useful in assisting clinical interventions related to retinal disorders.