Diabetic Retinopathy is a retinal disease because of high blood sugar. Early detection is necessary to maintain our vision. Small and large retinal hemorrhages can be detected using feature classification by extracting features from fundus images using filters bank outputs. Image level interpretation can be done by using difference of first and second order derivatives of Gaussian filters from green channel result. Finally a set of extracted features are selected with the help of filter approach and wrapper approach. k-NN classifier is used for final feature classification.