Diabetic retinopathy (DR) is the main cause of retinal damage due to fluid leakage from blood vessels. Automated diagnostic systems assist the ophthalmologists practice manual lesion detection techniques which are tedious and time-consuming. A Diabetic Retinopathy Lesion Discrimination (DRLD) model is proposed for abnormality identification followed by DR lesion detection based on identification of DR pathological symptoms. Shape, intensity and gray-level co-occurrence matrix (GLCM) features are extracted from the identified lesions, and exhaustive statistical analysis is performed for optimal feature selection. Overall accuracies of 97.9% and 91.5% are obtained using multi-layer perceptron neural network (MLPNN) and support vector machine (SVM) classifiers, respectively, for non-diseased versus diseased fundus image discrimination. MLPNN provides better performance for the fundus image discrimination approach, and further accuracy of 98.9% is obtained for DR lesion detection. When compared with other state-of-the-art techniques, the proposed approach provides better performance with significantly less computational complexity. A maximum accuracy improvement of 20.13% in fundus image discrimination and 5.90% in lesion categorization is achieved.