Detecting Hard Exudates In Retinal Fundus Images Using Convolutinal Neural Networks

Abstract

The main objective of this project is to detect Exudates in retinal fundus images using Convolutinal neural networks. Disorders in Retinal Images like Micro aneurysm, Hemorrhages', Hard Exudates, Soft Exudates, Macular Edema, Red lesions, Diabetic Retinopathy are likely to lead to severe visual loss Impairments. This work provides an automatic image processing techniques to diagnose Exudates in human eye and discussed various approaches used to detect Exudates in retinal images. Various publically available databases are listed and provide comparison between different approaches like SVM, KNN and CNN. Diabetic Retinopathy (DR) is the most essential causes of imaginative and prescient loss in diabetic patients. The most primary sign of DR is the presence of exudates, and detecting these in early screening is crucial in preventing vision loss. The automatic reputation of DR consisting of lesions, they are hard exudates (HEs), in fundus pix can make contributions to the diagnosis of this disease. On this take a look at, a fixed of functions from image regions are extracted and decided on the subset which quality discriminates between HEs and the retinal historical past. In proposed system, threshold based segmentation is used for extracting the features. After that, HOG (histogram of gradient), Classify the diseases using the convolutional Neural Network (CNN) classifier. The publicly available STARE of color fundus images was used for testing purposes and the values of sensitivity, specificity and accuracy were found as 96%, 98% and 99.68% respectively for the neural network based classification.