Skeletonization of Low-Quality Characters Based on Point Cloud Model

摘要

Skeletonization of low-quality Characters (LCs) is a very difficult problem. Since only detected contours (DCs) are known, existing methods focus on how to extract skeletons only from well located real contours (RCs), named real contour model (RCM), perform very badly. A new model, named point cloud model (PCM) is proposed to replace RCM in extracting skeletons for LCs. PCM can preserve more information for LCs and can obtain satisfied skeletons for LCs based on principal curves. The experimental results also show that our method proposed in this paper can obtain satisfied skeletons for LCs, especially in preserving topology and being consistent with the human perception even in serious quality reduction.