脱机手写体汉字切分算法分析-analysis of off-line handwritten chinese character segmentation algorithm.docxVIP

中南民族大学硕士学位论文 摘 要 脱机手写体汉字切分是脱机手写体汉字识别的基础，其准确率直接影响 识别的正确率，该问题的研究对提高识别的系统性能具有 重要的意义。本文 以脱机手写汉字为研究对象，对 汉字投影切分算法的相关问题进行了研究， 提出了改进的切分方案并对方案进行了实现 ，主要工作和创新如下： 1)在手写体汉字的书写过程中，会出现字符粘连现象， 该现象一直阻碍 着手写体汉字的切分准确率的提高。针对 上述问题，本文提出了基于多阈值 和多切分策略的间隙切分算法。 在阈值选取上，该算法采用多个的空隙阈值 分别进行切分，最终选取切分效果最优 的切分方案；在切分策略上，采取多 步切分策略，对过疏远或者过紧密的字符串分阶段进行处理。实验证明，较 传统的投影切分算法，该算法的切分准确率有 了一定的提高。 2) 在基于多阈值和多切分策略的间隙切分算法的基础上，本文通过对影 响切分准确性的若干因素进行研究，加入了标点符号切分块的标识、非规范 性笔画的修正、各切分块的块长统计等处理，提出了基于极小阈值和曲线拟 合的垂直投影汉字切分算法。该算法针对多阈值处理过程繁杂 且不够客观的 问题，使用极小阈值进行切分，并制定了相应的后期合并策略；针对 阈值法 寻找切分点的不确定性问题，在行切分和 字切分的再切分阶段，采用曲线拟 合方法寻找切分点。与一般的投影切分算法相比， 该方法在一定程度上提高 了切分正确率，并且适用性更广。 关键词：脱机手写；切分；投影；阈值；切分块 I 脱机手写体汉字切分算法研究 ABSTRACT Offline handwritten Chinese character segmentation is the foundation of offline handwritten Chinese character recognition, the accuracy of offline handwritten Chinese character segmentation directly affects the rate of recognition. Researches on offline handwritten Chinese character segmentation have very vital significance for improving the overall system performance of offline handwritten Chinese character recognition. This thesis takes offline handwritten Chinese character as the research object, makes studies of problems related Chinese characters projection segmentation, and makes corresponding improve at last. The main work and innovation as follows: In the process of writing Chinese characters, it may exit phenomenons that characters adhesion and characters overlapping, which could reduce the accuracy of offline handwritten Chinese character segmentation. In order to solve the problem, the thesis puts forward Clearance Project Segmentation Algorithm Based on multi-threshold and Multi-stage Cut. This algorithm uses multiple thresholds to do segmentation several times, and then selects the segmentation scheme which has the best segmentation result. Besides, the algorithm divides segmentation into several stages, deals with the estranged characters and close characters in different phases. Experiments show that, com