笔迹鉴别的字符予处理与匹配

Character Image Preprocessing and Matching for Writer Identification

笔迹鉴别多用匹配方法比较字符的书写风格,而字符图像的预处理和归一化对匹配是非常重要的。本文介绍笔迹鉴别的字符图像预处理和一种形状匹配方法。预处理主要介绍二值图像的噪声消除和归一化方法。噪声消除的方法是平滑、轮廓跟踪和填充。为保持字符中的书写特征,点阵的归一化是线性的,但字符位置和尺度的确定非常重要。本文给出了三种归一化方法:四边定界法、重心对准法和单边定界法,并在此基础上用图像匹配方法进行书写人识别的实验。匹配方法是通过距离变换快速实现的。实验结果表明,重心对准归一化最适合于笔迹鉴别问题,距离变换匹配得到的识别率也比较令人满意。

In writer identification(WI) ,writing styles are always compared by character match-ing ,while image preprocessing and normalization are very important for matching. In this pa-per, the techniques of preprocessing and shape matching for writer identification are presented. For preprocessing the noise elimination and size normalization are treated in this paper. The noise of character images is eliminated via smoothing,contour tracing and filling. The size nor-malization is linear in order to preserve the writing features. And the character position and scale are determined by three approaches ? square bounding, centroid alignment and single bounding. After normalization, writer recognition is implemented by character shape matching using distance transform(DT)which is computationally simple and distortion tolerant. The ex-perimental results show that centroid aligned normalization is the most suitable for writer iden-tification. The recognition rate is satisfactory which demonstrate the effectiveness of DT matching.