The off-line printed Yi character recognition system consists of four main modules, including character segmentation, feature extraction, feature compressing and dictionary matching. In this system, Yi characters' merger rules and anti-merger rules are summarized to increase the accuracy of character segmentation; 1024 dimensional features of peripheral direction contribution are used as statis-cal characteristics of Yi character with good discrimination between a large numbers of Yi similar characters. Feature compressing algorithm based on Karhunen-Loeve transformation and three-level dictionary -matching algorithm is also adopted to realize an off-line printed Yi character recognition system based on Windows. Its first recognition rates of samples are above 99.4% , which proved that all arithmetic used in this system are feasible and efficient.%脱机印刷体彝文文字识别系统包括字符分割、特征提取、特征压缩以及字典匹配四个主要模块,该系统利用总结出的彝文字符合并和反合并规则提高了字符分割准确率,采用1024维周边方向贡献度作为彝文字符统计特征,对彝文中存在的大量相似字符具有良好的区分能力.系统还采用基于KL变换的特征压缩算法和三级字典快速匹配算法,最终实现了一个基于Windows平台的脱机印刷体彝文识别平台,该平台对样本的一次识别率在99.4％以上.实验结果表明这些方法是可行的和高效的.
展开▼
