Skeleton is an important representation of objects. In this paper, an approach of extracting the skeleton based on the regions labeled from a gray image is developed. It specifies the complementary definitions of the ridge points, utilizes both contour and region information of an object, and uses hierarchical processing strategy. The multi-scale skeleton of an object can be robustly extracted, which is suitable for both regular and irregular objects. It is connected and single-pixel wide, and keeps the topological properties of the original object image. This approach is applied to real images, and the skeletons detected are consistent with visual apperception.