Abstract: In this paper, we represent a novel method for mesh simplification. Compared with
conventional methods that are based only on local features, our method exerts visual saliency feature
as the weight to sample points during simplification, hence has better performance on persevering
visual features. Firstly, the method computes the local feature values by the relationship between
Voronoi-poles and the sample points. Then the global visual saliency features are computed by
overlaying the Difference of Gaussian of local feature values get under different scales. Finally, the
computed global visual saliency features are used as the weight to the quadric error metrics, in order
to hold the positions where have of higher global visual saliency during iteration of the mesh
simplification. The simplified results show that our algorithm not only gets better visual feature
preservation but also smaller qualified errors comparing to state of the art methods.

Key words: Voronoi poles, saliency detection, mesh simplification