For　lubrication,　energy　saving　and　emission　reduction,　some　＇valley＇　textures　are　required　for　the　finishing　cylinder　liner　inner　surface　to　store　oil,　but　some　difficulties　are　brought　by　this　hierarchical　functional　surface　during　the　surface　quality　assessment.　The　reference　which　is　extracted　by　traditional　filtering　method　and　the　baseline　fitting　method　often　be　lowered　by　the　＇valley　＇　and　eventually　deviate　from　the　ideal　surface,　resulting　in　the　distortion　of　the　reference　plane,　making　the　subsequent　parameters　can　not　be　assessed　precisely.　Nowadays,　to　solve　this　problem,　robust　Gaussian　regression　filtering　is　highly　recognized,　but　eventually　being　too　time-consuming　and　can　not　be　promoted.　Therefore,　a　new　robust　regression　fitting　algorithm　is　proposed,　which　is　based　on　quadratic　function　and　trigonometric　function,　combined　with　the　Levenberg-Marquardt　algorithm＇s　minimizes　the　maximum　deviation　(least　absolute　value,　L∞)　regression　method,　making　lower　the　impact　of　outlier　on　the　hierarchical　functional　surface,　then,　reconstruct　the　datum　quickly　and　perfectly　to　be　achieved.　The　simulation　results　show　that　compared　with　least　squares　(Ls),　robustness　is　greatly　improved　by　the　proposed　method;　compared　with　Gaussian　regression　filter,　it　has　the　same　robustness,　but　the　speed　increases　nearly　105　times,　and　the　deviation　between　the　surface　in　three-dimensional　amplitude　parameters　and　function　parameters　with　the　true　value　do　not　exceed　2%　in　the　ISO　25178-2　standard.　Finally,　the　experiments　are　conducted　on　the　actual　cylinder　liner　inner　wall　surface,　and　the　results　show　that　the　new　method　is　more　robust　comparing　with　Ls,　more　time-saving　comparing　with　Gaussian　regression　filter,　and　it　has　high　precision　and　high　efficiency　during　extracting　datum.　©2014　Journal　of　Mechanical　Engineering