Non-Local　Means　(NLM)　denoising　algorithm　uses　Gaussian　weighted　Euclidean　distance　to　measure　the　similarity　between　neighborhoods,　but　when　the　image　noise　level　is　high,　the　distance　measurement　method　is　not　well　filtered　noise.　In　this　paper,　an　improved　nonlocal　mean　filter　image　denoising　algorithm　is　designed　by　analyzing　the　shortcomings　of　Gaussian　weighted　Euclidean　distance　in　measuring　neighborhood　similarity.　The　gradient　information　of　a　filtered　image　is　used　and　applied　to　the　Gaussian　weighting　coefficient　to　achieve　the　purpose　of　adaptive　filtering.　The　experimental　results　show　that　the　algorithm　has　better　denoising　effect　than　traditional　algorithm,　especially　when　the　noise　level　is　high,　it　has　high　peak　signal　to　noise　ratio.