Abstract: Aiming at the requirement of keeping the detail features during the simplification of triangle mesh models, awe proposed an optimized algorithm, awhich classified the vertex attribute based on discrete curvature estimation, aand it then combined the region growing method to adaptively determine the topological structure of the 3D models．We organized all the vertices by attributes, aand for each region we applied adaptive K-ring disk searching method to iteratively choose the similar attribute as a new seed．Our method started from curvature extrema points and searched concave and convex topological area by using effective region growing methods；it thus addressed models'features and topological structures．In the end a series of experiments show its efficiency．