Diabetes is a severe eye disease affecting various parts of the body including the eye. If diabetes is not properly controlled for more than 8 years, the retinal blood vessels in the human eye are damaged. The eye pathologies due to the damage of the blood vessel in the retina cause diabetic retinopathy which is a severe eye disease leading to vision loss. The major symptoms for this disease are the exudates which are due to the leakage of lipids and proteins from the damaged blood vessels. To prevent the people suffering by diabetic retinopathy from the vision loss, this disease should be diagnosed in an earlier stage. The chemical solutions applied into the patient's eye for capturing dilated retinal images cause irritation to the human eye. So, the nondilated retinal images are collected for diagnosis from the STARE web database. The input retinal images are preprocessed and segmented by using Fuzzy C-means segmentation technique. The features are extracted by Gabor filter and the features are compared by Speeded UP Robust Features (SURF) and Scale Invariant Feature Transform SIFT algorithms. Gaussian Mixture Model (GMM) classifier is used to identify whether the input retinal images are normal or abnormal images. The average classifier accuracy of GMM classifier in this research work is found to be 97.78%.