BP神经网络下的毛笔字帖字符切分方法研究  

Character Segmentation of Calligraphy Based on Back Propagation Neural Network

作　　者：朱训龙 袁国良[1] 周鹏 ZHU Xun-long;YUAN Guo-liang;ZHOU Peng(College of Information Engineering,Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《计算机仿真》2021年第10期497-501,共5页Computer Simulation

摘　　要：为了能够解决传统汉字切分方法中对大小不一和连笔的书法字体切分不准确的问题,提出了一种基于反向传播神经网络的毛笔字帖书法字体切分算法。算法在基于极小阈值和曲线拟合的垂直投影汉字切分的基础上,结合决定系数进行数据采样,将采样得到的数据进行预处理,输入到反向传播神经网络中进行分类。实验结果表明,相对于基于极小阈值和曲线拟合的垂直投影汉字切分方法,上述算法对含有大小不一和连笔现象的书法字体的切分是有效的,从而使书法字体的切分准确率有较大的提高。In order to solve the problem of inaccurate segmentation of calligraphy fonts with different sizes and continuous strokes in traditional Chinese character segmentation methods,a calligraphy font segmentation algorithm of brush calligraphy based on back propagation neural network is proposed.This algorithm combines coefficient of determination with minimum threshold and curve-fitting to sample data.The sampled data are preprocessed and input into back propagation neural network for classification.Experimental results show that this algorithm is effective for different size and adhesion characters,and it can improve the calligraphic segmentation.

关 键 词：反向传播神经网络 投影法 曲线拟合 决定系数 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]