部分级联特征的离线手写体汉字识别方法

Offline Hand-Written Chinese Character Recognition Based on Partial Cascade Feature

针对汉字类别多、风格多等识别难点,提出了一种基于LS-SVM的部分级联特征分类的离线手写体识别方法.具体包括霍夫空间加权采样和局部二值分布直方图两种新的特征提取算法,其可将任意大小的图像映射到固定长度的特征向量上,克服了已有特征提取算法的需要归一化、对笔画密度分布敏感等缺点;提出了基于部分级联特征的分类方式;提出了常见多分类方式的类别与正确率的关系,并给出了相应的数学证明.

A method for offiine hand-written Chinese character recognition is proposed based on partial cascade feature classification, which is of much research value and highly innovative. Two feature extracting algorithms are proposed as follows： weighted Low Threshold Hough Space Sampling（wHHS） and Histogram of Local Binary Distribution（HLBD）. These algorithms can map images of various sizes into vectors with fixed dimension, but eliminate the disadvantages of existing algorithms, which has high sensitivity of the distribution of strokes destiny, and demand uniformization. A strategy of classification based on partial cascade feature is proposed and the relationship between number of category for classification and accuracy is put forward with the corresponding mathematical proof.