用于相似字识别的手写汉字特征优化方法

A feature optimization method for similar handwritten Chinese character recognition

针对常用的手写汉字特征提取方法不利于后续线性区分分析(LDA)特征变换中发现相似汉字的细微区分信息,通过将传统的手写汉字特征提取和LDA变换表述为像素级特征的二维特征矩阵优化问题,并利用二维线性区分分析(2DLDA)变换进行手写汉字特征矩阵的优化,提出了一种用于手写相似汉字识别的特征优化方法.该方法可以避免高维像素级特征向量利用LDA变换进行优化中的散度矩阵奇异性问题.对手写相似汉字的识别实验表明,相对于传统的方法,经过所提方法优化的梯度特征,识别错误率可以降低48.86%,验证了方法的有效性.

In popular handwritten Chinese character feature extraction methods, feature vectors are mainly construeted by using sub-region partition and summing up the pixel feature inside sub-regions. This method is not conducive to find the subtle discriminative information among similar handwritten Chinese characters in subsequent Linear Discriminant Analysis （LDA） transformation. By representing the traditional feature extraction and LDA transformation as an optimization task on a two-dimensional feature matrix of the pixel level features, and using 2DLDA for the optimization of the handwritten Chinese character feature matrix, a new feature optimization method for similar handwritten Chinese character recognition was proposed. The proposed method can avoid the singularity problem of scatter matrices when using the LDA directly to optimize the high dimensional pixel-level feature vector. The experimental results on similar handwritten Chinese character recognition indicated that the optimized gradient features using the proposed method can effectively improve the recognition performance, and the recognition error rate reduction reaches 48.86% compared to the traditional method, showing the effectiveness of the proposed approach.