几何信息与SIFT特征相结合的特定人手写关键词检测

Specific handwritten keyword spotting using geometric information and SIFT feature

中文汉字类别繁多,书写随意性大,使得中文的手写体关键词检测具有很大的挑战性。提出一种基于文字几何信息和SIFT特征相结合的手写体关键词检测方法,通过计算文本图像特征的匹配度来检测特定书写人的手写关键词。尺度不变特征转换(scale invariance feature transform,SIFT)局部特征具有良好的稳定性和独特性,既能适应同一书写人手写汉字的差异,又能区分不同书写人的书写笔迹。结合文字的几何信息,通过滑动窗口和最大团查找方法可以有效地删除误匹配点,极大地提高关键词检测的成功率。对大量手写体文本图像的实验结果表明,该方法能够有效检测同一书写人的相同关键词,具有较高的召回率和准确率。

Large variety of Chinese characters and handwriting styles leads to a big challenge for keyword spotting in Chinese handwritten documents. A new method combining the character geometric information and SIFT feature is proposed for detecting handwritten keywords of specific handwritten. It is proven that SIFT is a stable and distinctive local feature,which can perform well in distinguishing different handwriting styles. Combined with character geometric information and maximum clique matching,the proposed method can effectively remove miss-matching feature points and improve the precision rate of detection. Experimental results in handwriting document images show that the method can efficiently detect keywords of particular writers and remain high recall rate and high precision rate.