Research On Skeletonization Of Low-quality Chinese Characters

A difficult problem of skeletonization has been discussed in this paper: skeletonization of low-quality Chinese Characters (LCCs). Since LCCs are affected by a variety of low-quality factors, most of existing methods can not extract skeletons for consistent with human perceptions and meeting the "good" skeletons standard.In this paper, we propose a new model for LCCs firstly, named point cloud model (PCM), for skeletonization of LCCs. PCM can fully preserve the existing underlying information of characters, it not only changes contour feature into stroke point characteristic, but also provides an unified and feasible model for LCCs and the general characters. Based on PCM, this paper considers the skeletonization of LCCs as an optimal problem with two steps: firstly, find out the primary skeletons which are consistent with the original topology of characters and composed by some line segments, named primary skeleton segments (PSSs); and then, connect the PSSs, to obtain the skeleton of LCCs which conform with human vision and maintain the original topology of the LCCs.This paper integrates feature analysis, clustering, optimization, and random field theory together to present a new method for skeletonization of LCCs. In this method, LCCs are considered as point clouds firstly. And then uses of principal component analysis (PCA) for dimension reduction. This paper proposes the incre- mental generalized k means clustering (IGKMC) for extracting PSSs. Finally, we consider the problem for connection skeleton as labeling problem, and can handle the problem by optimization methods, the PSSs are connected under the framework of High-level Markov model (HLMRF) in this paper. Based on HLMRF, prior is added to define the energy between the primary skeleton segments (such as angle difference, shift difference, vertex distance), to obtain the optimal solution of the labeling problem.Our main contributions in this paper are: we develop the PCM for LCCs, and then bases on the model, we put forward a new scheme to extract skeleton for LCCs. Experiments show this new method for LCCs skeletonization can provide "good" skeletons even in serious quality reduction cases. In addition, this paper establishes a new model and theoretical framework for LCCs skeletonization, and also provides a new idea for LCCs skeletonization. We believe that this work can inspire new approaches and new ideas for studying on LCCs skeletonization.