面向汉字书写质量评价的 Gabor 特征提取方法-《计算机技术与发展》

文章信息/Info

Title:: Gabor-based Feature Extraction towards Chinese Character Writing Quality Evaluation

作者:: 闫文耀¹; 郭铭涛²; 王志晓²; 3; 张九龙²; 1. 延安大学西安创新学院数据科学与计算机学院,陕西西安 710100; 2. 西安理工大学计算机科学与工程学院,陕西西安 710048; 3. 陕西省网络计算与安全技术重点实验室,陕西西安 710048

Author(s):: YAN Wen-yao¹; GUO Ming-tao²; WANG Zhi-xiao²; 3; ZHANG Jiu-long²; 1. Xi’an Innovation College of Yan’an University, School of Data Science and Computer Science, Xi’an 710100, China;? 2. School of Computer Science and Engineering,Xi’an University of Technology, Xi’an 710048,China;? 3. Shaanxi Key Lab of Network Computing and Security Technology, Xi’an 710048,China

Keywords:: Chinese character; writing quality evaluation; texture feature; Gabor transform; SVM

摘要:: 汉字书写质量的智能化自动评价具有广泛的应用前景。已有的汉字书写质量评价大多基于部首分割和特征判决,此类评价方法的一个关键环节在于对汉字分割的精度要求高。但是,真实手写字体通常存在连笔、落笔力度不均匀、个性化的书写风格等问题,这导致普通书写的字存在分割困难的问题。针对上述问题,提出一种基于图像纹理的书写质量评价方法。该方法将书写线条的匀称清晰程度以及书写风格的一致性作为有效判据,具体采用 Gabor 变换对书写样本的图像特征进行提取,最终采用支持向量机的统计学习方法对书写质量进行有效评判。在 CHAED 字库集等多个真实数据集上的实验展示了该方法是有效且准确的,其优势在于无需对字体进行分割,且其全局性特征的提取过程计算代价较小。

Abstract:: The intelligent automatic evaluation of the writing quality of Chinese characters has a wide application prospect. Most of the existing evaluation methods of writing quality of Chinese characters are based on radical segmentation and feature judgment. A key part of such evaluation methods? ?is the high requirement of Chinese character segmentation precision. However,real handwriting usually has some problems such as connecting strokes, uneven strokes and personalized writing style, which leads to the difficulty in separating ordinary handwriting. Therefore,we propose a writing quality evaluation method based on image texture. In this method,the symmetry and clarity of writing lines and the consistency of writing style are taken as effective criteria,specifically with Gabor transform to extract the image features of writing samples,and finally statistical learning method of support vector machine is adopted to effectively evaluate the writing quality. Experiments on multiple real data sets,such as CHAED word set,show that the proposed method is effective and accurate.Its advantage lies in that font segmentation is not required, and the calculation cost of global feature extraction process is small.