摘要
手写体汉字离线识别系统实际应用的关键问题之一是如何维护一个既有足够代表性又不至于过于庞大的汉字字库.本文提出了用并行遗传算法有效地从庞大的不同风格手写体汉字全集中选择出样本字,构建识别意义上最优的压缩样本字库的方法.采用ETL882字库和文献[1]中的识别算法所做的实验,证明了这一方法的有效性.
One critical problem influencing the real world application of off-line handwritten Chinese character recognition technique is how to maintain an appropriate sample character library: not too large for the efficiency of computation, and meanwhile not too small for the accuracy of recognition. The compromise lies in that how well a selected sample library represents the assumed total set of input characters. In this paper, an optimal compressed sample library is defined, and PGA is used to achieve this goal. Preliminary test on ETL8B2 using recognition algorithms provided by reference[1], attests the validity of this method.
出处
《南开大学学报(自然科学版)》
CAS
CSCD
北大核心
2003年第3期62-66,共5页
Acta Scientiarum Naturalium Universitatis Nankaiensis
基金
天津市自然科学基金资助项目(983600511)
关键词
离线手写汉字识别
遗传算法
并行遗传算法
样本库压缩
offline handwritten-chinese character recognition
genetic algorithms
parallel algorithms