Accurate　retinal　Image　registration　is　essential　to　monitor　and　track　the　progress　of　various　diseases.　Since　it　is　low　quality　images　on　Non-mydriatic　or　with　the　disease.　Then　vascular　structure　will　become　less　clear.　It　becomes　more　difficult　for　the　general　registration　methods.　In　this　paper,　a　novel　feature　based　retinal　image　registration　method　is　proposed　to　solve　this　problem.　SIFT(Scale　Invariant　Feature　Transform)　as　key-point　is　extracted.　Best-Bin-First　(BBF)　and　keypoints＇　orientations　are　applied　to　identify　the　corresponding　features　between　two　images.　Affine　model　and　quadric　model　are　used　to　register　two　images,　which　is　based　on　strategy　of　coarse　to　fine　registration.　Experimental　results　show　that　the　method　is　robust　and　efficient.