Evaluation of Spatial Filtering Techniques in Retinal Fundus Images

Abstract

The denoising of the fundus images is an essential pre-processing step in glaucoma diagnosis to ensure sufficient quality for the Computer Aided Diagnosing (CAD) system. In this paper, we present an evaluation approach for different denoising filters of eye fundus images that suffer from two different types of noises (Gaussian noise and Salt & Pepper noise), which had been applied to the retinal images and then various Spatial filtering techniques like linear (Gaussian, mean), nonlinear filtering (median) and adaptive filtering have been implemented to three types of images (original images, images with salt and pepper noise and images with Gaussian noise) and their performance are compared to each other based on evaluation parameters: Mean Squared Error (MSE), Peak Signal Noise Ratio (PSNR) and Structural Similarity (SSIM). The results showed that the adaptive median filter has the best performance in salt & paper noise and the adaptive filter has the best performance for Gaussian noise, but their performance is close to each other. In conclusion, six spatial filters applied to RIM-ONE fundus image database and found that, the adaptive median filter has the best performance compared to other filters to remove these noises and increase the quality of the resulting images, which can be implemented to the CAD system.