An　automatic　dimension　measurement　method　of　the　scratch　images　is　proposed　based　on　the　level　set　and　the　anisotropic　diffusion.　The　segmentation　and　dimension　measurement　are　conducted　on　scratch　images　that　are　obtained　by　optical　microscope.　In　the　segmentation,　the　distance　regularized　level　set　evolution　(DRLSE)　is　used　to　segment　the　scratch.　The　anisotropic　diffusion　method　is　utilized　instead　of　Gaussian　filtering　as　the　de-noising　method　of　the　edge　indicating　function.　The　initial　level　set　is　manually　set　and　evolved　to　the　whole　scratch　boundaries.　The　contour　coordinates　are　achieved　by　the　subpixel　adjustment.　In　the　dimension　measurement,　the　scratch　boundaries　are　used　to　measure　the　width　distribution　and　scratch　length.　Meanwhile,　the　average　scratch　width　and　scratch　area　are　obtained.　The　influence　of　evolution　parameters　on　measurement　results　are　discussed　by　experiment.　Under　the　optimal　parameters　of　evolution,　the　relative　measurement　errors　of　the　average　scratch　width　and　scratch　area　are　less　than　0.06%　and　0.1%.　Compared　with　the　traditional　DRLSE,　experimental　results　show　that　the　proposed　method　has　advantages　of　precision　and　accuracy　in　the　average　scratch　width　and　scratch　area.　©　2020,　Science　Press.　All　right　reserved.