The　majority　of　image　denoising　algorithms　assume　that　the　noise　is　evenly　distributed　white　Gaussian　noise,　however,　the　image　noise　which　collected　in　real　scenes　is　more　complex.　In　this　paper,　we　propose　a　real　image　denoising　method　based　on　pixel-level　noise　estimation.　The　method　is　improved　on　the　basis　of　the　block-matching　and　3D　filtering　(BM3D)　image　denoising　algorithm.　The　noise　estimation　algorithm　further　introduces　pixel-level　non-local　self　similarity　(NSS)　on　the　basis　of　patch-level　NSS　prior.　After　detecting　the　flatness　of　the　image　block,　the　relevant　parameters　are　adaptively　adjusted,　finally　the　noise　estimation　algorithm　and　the　image　denoising　algorithm　are　combined　block　by　block.　Experiments　show　that　this　noise　estimation　method　greatly　reduce　the　required　processing　time　while　ensuring　the　accuracy　of　noise　estimation.　The　image　denoising　effect　has　certain　superiority　compared　with　other　classic　traditional　denoising　algorithms.　©　2022　ACM.