一种LLE降维表示的汉字字体识别方法

A Chinese Font Recognition Method Based on LLE Dimension Reduction

为解决汉字字体的识别问题,在图片字体灰度值像素点特征上使用局部线性嵌入(Locally Linear Embedding,LLE)降维算法进行流形分析,在获得的子空间内利用相似字体在子流形空间近邻的这一特性结合最邻近分类算法,构建了一种汉字字体识别方法。在四种常用字体数据集上进行广泛实验,该方法对于黑体、楷体、圆体、装饰体的平均识别率分别为74.32%、77.53%、79.52%、90.25%。实验结果表明:该方法对字体的识别整体来说是有效的。

In order to solve the problem of Chinese font recognition,this paper adopts the Locally Linear Embedding(LLE)dimension reduction algorithm to perform manifold analysis on the pixel features of the image font gray value.In the obtained subspace,the characteristics of similar fonts in the submanifold space and the nearest neighbour classification algorithm are used.A method of Chinese font recognition is constructed.Extensive experiments were carried out on four common font data sets,and the average recognition rates of boldface,italic,round and decorative fonts were 74.32%,77.53%,79.52%and 90.25%,respectively.Experimental results show that this method is effective for font recognition.