摘要
根据 Hamming神经网络的工作原理提出了一种可用于手写体数字识别的结构简单灵活的电流型可编程局部结构特征提取电路 .该特征提取器的模板不仅是可编程的 ,可根据不同的需要来随时更改模板的内容以适应不同的情况 ,而且其特征合并的方式也是可配置的 ,可根据不同的需要把所需的特征合并成不同的特征类别 .对于该可编程特征的电路模拟以及用单层多晶、双层金属的 1 .2μm数字 CMOS工艺所制作的实验芯片的测试表明 ,该电路能很好地完成特征提取的功能 .
A smart current\|mode programmable local feature extractor for handwritten digi t classification is put forward based on the principle of Hamming neural network . The templates can be programmable and their contents can be refr eshed according to different situation, and the way of features merging is also r econfigurable so as to merge the different features into different feature class es. A prototype chip of the feature extractor has been implemented in 1.2 micron singl e poly, double metal digital CMOS technology and both the simulated and measured results prove that the extractor can realize the feature extraction well.
基金
国家自然科学基金资助项目! ( 6963 60 3 0 )&&
