Identification of noise in the fundus images

Analysis of the tiny retinal vasculatures in retinal fundus images becomes difficult due to very low and varied contrast between the retinal vasculature and the background. Fundus fluorescein angiogram overcomes these problems and provides an excellent visualization of the retinal vasculature; however it is an invasive procedure requiring injection of contrasting agents. Further investigation of the RETICA method, a non-invasive method of image enhancement developed earlier, is reported in this paper. It was found that noise is present in the Retinex image. Thus, the identification of the noise in the Retinex image and its removal has been the focus of this research paper. The method used to identify noise is based on adaptive wiener filters (additive, multiplicative, and additive plus multiplicative filters) and the fundus model image and real fundus images are applied to these filters. It is observed that retinal fundus images contained both additive and multiplicative noise. The noise is reduced by using adaptive wiener filter (additive plus multiplicative adaptive wiener filter) based method which resulted in 2.84db an improvement in PSNR.

I. Introduction

Eye screening is vital in detection of diabetic retinopathy. There are five stages in diabetic retinopathy (DR) ranging from normal (No-DR), mild DR, moderate DR, severe non proliferative DR (NPDR) to PDR. The PDR stage is where there is a total loss of vision [1]. Haemorrhages, exudates and changes in the veins are some of the pathologies that, when present, characterise these DR classifications [2]. It was reported in a research carried out recently on the analysis of images of the fundus that as the severity level of DR advances, the fovea avascular zone size increases. The FAZ is observable in colour fundus images and in fundus fluorescein angiograms (FFA) [3].