An Improved Accuracy Rate in Microaneurysms Detection in Retinal Fundus Images Using Non-local Mean Filter

Abstract

Microaneurysms (MA) detection in diabetic patients is very important as it’s the first phase in grading the Diabetic Retinopathy disease through retinal fundus images. This paper presented a method to improve the accuracy rate in MA detection using a Non Local Mean Filter (NLMF). MA is one of the small retinal features in fundus images and it is a very challenging task to segment it since it will be merged with other retinal parts and features. The noise has affected the overall performance of MA detection methods, hence this paper has proposed NLMF. The methodology starts firstly, the RGB fundus images are converted into green channel images, then the mask is generated for the green channel image and then the background noise is reduced by the NLMF approach. Secondly, the hidden features are extracted using the contrast-limited adaptive histogram equalization further if there is exists any noise, that is removed by the 2D Gaussian low pass filter. Finally, the filtered images are converted into binary images and then MAs are extracted using morphological top-hat transform. The proposed MA detection was examined using the available publicly accessible dataset such as e-Ophtha, Retinopathy Online Challenge (ROC) DIARETDB0 and DIARETDB1. The presented MA detection method has produced better accuracy with other state of the art methods.