摘要
针对手写阿拉伯文100类变体字符中相似字多、书写粘连等识别难点,提出一种基于部件分解和自适应融合的识别算法.首先,根据结构规则建立字符的部件模型,将字符分解为主体、附加和点三类部件,并结合连笔分析获得较鲁棒性的部件描述;然后,针对各类部件的不同特点设计相应的特征抽取和分类器,通过部件匹配来检测和辨识相似字间的微小差异;最后,利用并改进D-S证据理论对多个部件进行融合,通过分析部件的匹配度分布建立一种实时的融合权重计算方法,并基于所得权重提出证据的折扣方案,从而实现自适应融合,以提升字符识别效果.实验证明该算法较现有经典算法在识别率和稳定性方面均有明显提高.
For 100 Arabic characters,a handwritten recognition algorithm based on radical decomposition and self-adaptive fusion is proposed.Firstly,the radical model is established by decomposing the Arabic character as three types of radicals： main,affix and dot.According to the analysis of connected strokes,a robust radical description is obtained.Secondly,different feature extractions and classifications are designed for various types of radicals,so that every radical is matched to detect and identify slight differences between similarities.Finally,a multi-radical fusion scheme based on the D-S evidence theory is developed.A new method to estimate the fusion coefficient is also proposed according to the confidence distribution of radicals.With the corresponding discounted mass refined based on the coefficient,all radicals can be self-adaptively fused to enhance character recognition effect.Analyses and experiments show that the proposed method can lead to a better performance than the present traditional algorithms.
出处
《西安电子科技大学学报》
EI
CAS
CSCD
北大核心
2012年第6期16-21,77,共7页
Journal of Xidian University
基金
国家自然科学基金资助项目(60872141)
中央高校基本科研业务费专项资金资助项目(K50510010007)
华为科技基金资助项目(HITC2011023)
关键词
手写文字识别
阿拉伯语
信息融合
自适应
证据理论
handwriting recognition
Arabic language
information fusion
self-adaptive
evidence theory
