【作者】 朱学芳； 邹文豪； 朱鹏；

【Author】 ZHU Xuefang ZOU Wenhao ZHU Peng Department of Information Management,Institute of Multimedia Info Processing, Nanjing University,Nanjing,210093,China

【机构】 南京大学信息管理系、多媒体信息处理研究所；

【摘要】 本文叙述了基于Gabor函数滤波器提取汉字版面纹理特征以及通过一系列预处理、求取待识别字体滤波后图像与已知字体滤波后图像相似度以进行汉字字体识别的方法。实验分析输入汉字图像的各属性值,获得了Gabor函数的各自由参数,得到了不同字号情况下的楷体,黑体,宋体,隶书四种汉字字体的应用特征图协方差,特征图方差,直方图协方差,直方图方差方法的识别结果,分析表明应用直方图方差法对较大号字体的识别率较高,正确识别率在96%～100%之间。更多还原

【Abstract】 Experiments on Chinese font recognition are investigated in the paper.The original input images are from the books,magazines,etc.,printed in Kai,Hei,Song and Li fonts of Chinese characters with seven different sizes.The parameters of filters based on bi-dimensional Gabor function are optimized firstly.Then the normalized input images are used for the extraction of texture feature after the removal of some noises.Finally,the experimental results are obtained by using the four kinds of similarity comparing methods,i.e.variations and co-variations of feature images and their histograms respectively,which shows that the bigger fonts have higher correct recognition rate than the smaller ones. They are between 96%and 100%for the four typical kinds fonts of Chinese characters.更多还原