基于Zernike矩的字符特征选择

Choose of Character Feature Based on Zernike Moment

对从工业现场提取的15类字符分别计算从2阶到14阶的Zernike矩模值,根据类内距离极小化和类间距离极大化的判别标准对Zernike矩特征值进行分析,最终选择出18个Zernike矩进行进一步归一化后作为字符特征输入到神经网络字符识别系统.实验表明,该组特征对旋转字符的正确识别率达99%.该方法选择字符的输入模式特征对旋转字符具有较高的识别率并可广泛应用于其它需要识别旋转图形的特征选择.

The fifteen classes of characters are obtained from industry spot and their Zernike moment amplitudes from the 2nd to 14th order are calculated. According to the criterions of the minimum intra-cluster distance and the maximums inter-clusters distance, eighteen Zernike moments are chosen after analyzing the feature values. After further normalizing the Zernike moment, the eighteen feature values are input to the neural network character recognition system, the experiment results demonstrate that the chosen features can obtain 99% correct recognition rate to the rotation characters, The high recognition rate can be obtained and the method can be applied to feature choice of other rotation images.