摘要
脱机手写体识别是字符识别中的难点之一,日文中的平假名类似于中文的手写体草书。为解决该问题,首先,针对日文平假名字符的特点提出了一种基于网格的外围特征提取方法,其次,考虑到了不同特征的分类能力的差异性,提出了一种基于支持向量机的多特征融合的识别方法,提高了识别率。最后,针对日文车牌中的6 735个平假名样本和4 145个数样本字进行了识别实验。实验结果表明,该方法的识别率可达98%左右,优于距离分类器及神经元网络的方法,具有实际应用的价值。
Hiragana is similar to Chinese cursive writing which is a difficult topic in character recognition field. To solve this problem, firstly, a novel feature extracting method for hiragana characters based on mesh peripheral feature was presented. Secondly, a recognition method based on multi-feature fusion and support vector machine was designed to promote recognition rate. Finally, experiments were conducted for 6 735 hiragana samples and 4 145 numerals from the Japanese vehicle plate. The experiment result of recognition rate is about 98% which has a better performance than distance classification and neural network.
出处
《实验室研究与探索》
CAS
北大核心
2009年第12期27-30,共4页
Research and Exploration In Laboratory
关键词
特征提取
支持向量机
特征融合
平假名
字符识别
车牌识别
feature extraction
support vector machine
multi-feature fusion
hiragana
character recognition
vehicle plate recognition
