Abstract: Most of the existing feature detection methods are differential geometry based,sensitive to noise,run slowly and cannot handle blend features very well.To solve these problems,a sparsity optimization based mesh feature detection method is proposed in this paper.This approach mainly consists of three procedures:first,the mesh is smoothed by a Laplacian energy function restrained by a l₁-norm sparsity term and a l₂-norm error term;second,initial feature points are extracted according to the moving distances;last,a post-processing is performed on the extracted feature points so that these features look better.In our approach,the l₁-norm is used to penalize the number of points moved and the l₂-norm is applied to control the degradation of the smoothed model.This method is easy to implement,and it can not only handle sharp features and weak features,but also deal with blend features.Compared with differential geometry based methods,the results show that the proposed method is simple,effective,and fast.Moreover,the extracted feature lines are superior to other methods.