Abstract: Extraction of skeleton of polygonal shape will be seriously affected by the noise on the shape boundaries.To overcome this problem, a novel feature preserving method for extracting approximated skeleton of the planar polygon is presented.First, prominent feature points are detected by the watershed algorithm which is resistant to the shape noise.Then these prominent points are used as leaf nodes to guide the extraction of skeleton from the shape.Therefore, the skeleton can be effectively extracted without redundant branches.Alternatively, the user can also specify the leaf nodes at branch points of the skeleton to obtain a better and more practically useful skeleton.Experimental results show that our method is easy to implement and can be applied to many applications in computer graphics including computer animation and shape retrieval.