Diabetic retinopathy (DR) is a serious complication of diabetes mellitus and the commonest cause of blindness in the working population of the world. Automated early detection of DR from the fundus images could limit the severity of the disease and assist ophthalmologists in more efficient treating. The aim of this work is to develop automated methods which can detect vascular disorders and non-perfused areas in color fundus images with an acceptable sensitivity and specificity.

The performance of six different preprocessing algorithms has been used and applied on ten color images to compare statistically, choosing the usefulness preprocessing method for segmentation stage. A vessel detection algorithm was studied and implemented on retinal images before and after preprocessing. Exudates were segmented using our morphological operations algorithm. The result of segmentation algorithm compared before and after using preprocessing methods on original images by calculation of sensitivity and specificity. To estimate the existence of capillary non-perfusion (CNP) area in fundus color images the two color components red and green were studied using calculation of texture specification values included mean of intensity, coefficient variation, smoothness, uniformity, entropy and forth moment. The area under Roc curve was calculated using a nonparametric estimate to measure the overall performance of the texture specifications (SPSS ver.13)

The average of sensitivity and specificity for vessel detection on original images are 69% and 97% respectively. The average of sensitivity rise to 74% for vessel detection by using local contrast enhancement on images as a preprocessing stage. Compared to a clinical reference standard, images containing exudates were detected with sensitivity 76% and specificity 98%. Statistically evaluation of texture specifications of CNP and normal areas in retinal images shows significant differences in some texture specification by P-value <0.05.

It was concluded using pre-processing techniques to correct and enhance colour founds images for diagnosis of lesions and interested areas are useful. The method introduced for segmentation of exudates shows good achievements so we suggest this technique could be developed using some algorithm improvements on more fundus images.