Skeletonization based on high-level Markov Random field

One essential challenge for some skeletonization methods are the skeletons are broken on the intersection regions or by noise. In order to connect the broken skeletons more reasonable and obtain good skeleton performance, we propose a method to repair the primary skeletons extracted by some recently methods. The primary skeletons are repaired using high-level Markov Random field (HLMRF) whose fetures are lines of primary skeletons. The final repaired skeletons are obtained by finding the optimal resolution of cost function defined by HLMRF. Experiments show the good performance of our method.