A High Performance Hybrid Technique of Microaneurysm Extraction Using Vessel Suppression and Connected Component Extraction

Abstract

Detection of microaneurysms in the early stage is a crucial step towards identifying non - proliferative diabetic retinopathy in diabetic patients. Micro aneurysms appear in the form of tiny red dots in fundus images near the blood vessels. Thus making it is very difficult to identify them through ordinary techniques. In this paper we propose a hybrid technique which uses vessel suppression in increase the performance and accuracy of microaneurysms detection. The entire process can be divided into four stages. First the pre-processing stage which enhances the contrast and eliminates noise by contrast limited histogram equalization and bilateral filter respectively. Then multiscale hessian transform based enhancement is performed which includes vessel extraction and suppression. Finally the microaneurysms are extracted using connect component extraction and classified based on features extracted. This technique eliminates the number of false positives detected at every stage thus improving the sensitivity of the system. The algorithm has been thoroughly tested on numerous images and it outperformed the existing counterparts.