Abstract: Traditional skeletonization algorithm based on the distance transform can not be used for skeleton recognition directly, and the connectivity property of the skeleton is not guaranteed. A novel skeletonization algorithm is presented, in which the whole skeleton is obtained by growing from the original skeleton seed one by one. In the growing process, the redundant skeleton branches are eliminated by the discrete curve evolution model, the visual branches remain completely, and the hierarchical control can be achieved easily. Examples have showed that the complexity of this algorithm is low, the connectivity of the skeleton is guaranteed, and the skeleton can represent the visual parts to satisfy human vision. The algorithm can be used in graphic recognition and shape measurement.