A new parallel 3D skeletonization algorithm based on sort ascending rechecking is proposed. A new method

named 8-6 trees

to identifying simple points is designed. Simple points can be identified by traversing 8-6 trees

without calculating Euler characteristic. Distance transform and parallel 6-subiteration thinning are combined by a new label rule and sort ascending rechecking

remedies the defects of usual sequential algorithms. The influence of noise is controlled by preserving convex vertexes. Skeleton generated by our algorithm is in accurate position

with correct connectivity and topology. This algorithm is applied to a large number of 3D images and good results were achieved. Furthermore

this algorithm can also be used to skeletonize 2D images.