一种用于脱机手写体汉字识别的多层流水线并行处理结构

A Parallel Processing Architecture for Off-line Handscript Chinese Characters Recognition Based on a Mu lti-layered Pipeline

对于脱机手写体汉字识别来说 ,速度一直是很重要的一个问题。而其中输入汉字特征向量和样本库中海量模板匹配占用了很多的识别时间 ,约为系统识别时间的 4/5,是提高汉字识别速度的瓶颈。研究了基于多层流水线 ,采用多组计算单元、多体存储器和多执行单元的并行处理结构 ,充分利用硬件特性 ,并行处理模板匹配。并在此基础之上 ,设计了手写体汉字识别专用芯片 ,进行了逻辑模拟 ,且用 FPGA实现了该设计 ,结果证明 ,采用此种方法使汉字识别速度大幅度提高。

For the off-line handscript Chinese characters recognition system, it is very important to improve the recognition speed.A parallel proc essing architecture is investigated in the paper,which consists of mul ti-layered pipeline,multiple sets of calculating units,multi-bank stor age,and multiple execution units.And based on this architecture,an AS IC for handscript Chinese characters recognition is designed,which is implemented in FPGA’s.The new circuit greatly increases the speed of off-line handscript Chinese character recognition.