Accurate detection of microaneurysm (MA) plays a very important role in early diagnosis of diabetic retinopathy. This paper presents a novel method based on the variation of local intensity for microaneurysms detection in retinal images. In contribution, proposed method use local rank transform effectively
for separation of MAs in the retinal images. Secondly, a novel blood vessels extraction method using a gradient of guided filter is proposed. Finally, selecting MA candidates by excluding vessels and post-processing techniques. Experiments were carried out on a three publicly available datasets E-Ophtha, Diaretdb1, and Messidor. The proposed method is effective regarding fast and accurately detecting MAs compared with the other state-of-the-art methods.