A comprehensive analysis for retinal image classification methods using real-time database

This paper proposes a comprehensive analysis to compare our two methods with some existing methods to prove the improvements of the proposed algorithm. Here, the comparative analysis is done by three phases like pre-processing, segmentation and classifier phase. For pre-processing, four noise removal filters like average, Laplacian, motion and unsharp are carried out to compare with Gaussian filter. The two segmentation algorithms like region growing and k-means segmentation are carried out to compare in proposed segmentation phase. For diabetic classification, the proposed classifier of Levenberg-Marquardt (LM) neural network against other four existing classifiers SCG-NN, adaptive neuro-fuzzy inference system and k-NN. Here, some different evaluation metrics such as PSNR, SSIM, sensitivity, specificity and accuracy are used to measure the performance. The test images are considered for performance analysis in real-time. The proposed approach obtained 97.5% in terms of accuracy for real-time database, which is high compared to existing techniques.