In　the　field　of　denoising　methods,　traditional　filters　are　still　widely　used　because　of　their　simple　programming　and　high　processing　efficiency.　However,　they　have　been　less　accepted　since　their　low　efficiency.　For　example,　Median　filters　cannot　denoise　Gaussian　noise　effectively,　while　mean　filters　would　lead　to　fuzziness　of　images　due　to　the　loss　of　lots　of　details.　Against　this　backgrounds,　we　proposed　a　two-stage　filtering　algorithm　based　on　local　similarity,　which　can　preserve　better　of　the　structural　details　and　contrast　of　images.　In　the　first　stage,　we　use　adaptive　median　filter　algorithm　to　filter　out　the　impact　of　impulse　noise;　And　then　we　do　the　second　image　denoising　according　to　the　principle　of　local　similarity　of　the　image.　Extensive　simulation　results　under　various　experimental　settings　demonstrated　the　effectiveness　of　the　proposed　algorithm.