【摘要】 针对手写湘西方块苗文的机器识别,提出一种基于PNN神经网络的分类识别方法。此方法是在经典的像素投影法基础上进行的改进,为了充分挖掘分类器的潜力,避免过多的特征矩阵维数导致的分类器过拟合问题,分别对文字图像投影到X轴和Y轴的像素进行了最大公约数化分组,以简化特征矩阵的维数,并使用提取到的多种像素分类方式下的特征矩阵,结合PNN神经网络进行训练和分类,验证了此种方法在不同像素分组下相对于原始投影法的识别率都有不同程度的提升。通过实验所有的像素均分情况,得到了识别率最佳的像素分组方式。更多还原

【Abstract】 Aiming at the machine recognition of handwritten Xiangxi Hmong character, a classification and recognition method based on PNN neural network was proposed. This method was improved on the basis of classical pixel projection method. In order to fully tap the potential of the classifier and avoid over-fitting of the classifier caused by too many dimensions of characteristic matrix, text images projected onto the pixel X axis and Y axis were classified based on great common divisors respectively to simplify the dimension of feature matrix. In addition, feature matrix extracted by multiple pixel classification methods, combined with the training and classification based on PNN neural network, has verified the improvement of recognition rate of the original projection method under different pixel groups. Through experimenting on the equalization of pixels, the pixel grouping method with the best recognition rate has been obtained.更多还原