基于Zernike矩与网格特征的两级神经网络字符识别系统

The two-level neural network character recognition system based on Zernike moment and grid features

从待识别字符中分别提取网格与Zernike矩特征,并将两种特征输入改进的BP神经网络进行比较,得知网格特征对随机噪声及笔划粗细变化不敏感,而Zernike矩特征对旋转不敏感,针对两种字符特征的缺点与优点,把以上两个神经网络进行联合,设计了一个两级神经网络字符识别系统,实验表明,基于不同特征输入的神经网络识别结果之间存在互补现象,上述两级神经网络分类器串级集成字符识别系统正确识别率达98%。

The grid feature and Zernike moment feature are abstracted from the recognized characters and the two features are input to the modified BP neural network. According to comparison it is known that the rigid feature is not sensitive to the random noise and the width change of strokes, and the Zernike feature is not sensitive to rotation. Combining the two features＇shortage and advantage, the two BP neural networks are integrated and a two- level neural network character recognition system is designed. The experiment results show that the complementary facts exist among the recognition results obtained by neural network based on different feature inputs, the character recognition system presented in this paper can obtain the correct recognition rate of 98%.