Notice of RetractionOne Radical-Based On-Line Chinese Character Recognition (OLCCR) System Using Support Vector Machine for Recognition of Radicals
Notice of RetractionOne Radical-Based On-Line Chinese Character Recognition (OLCCR) System Using Support Vector Machine for Recognition of Radicals
复制标题
DOI：
10.1109/icbbe.2007.146
复制
发表时间：
2007-07
期刊：
影响因子：
--
通讯作者：
Xinqiao Lv;Dongshan Huang;Enming Song;P. Li;Chunshan Wu
中科院分区：
文献类型：
--
作者：
Xinqiao Lv;Dongshan Huang;Enming Song;P. Li;Chunshan Wu
提出了一种基于支持向量机（SVM）的部首联机手写体汉字识别方法。在本系统中，输入的汉字经过预处理后，提取汉字的结构特征，并对每一段的中点进行水平和垂直方向的投影，以检测汉字的模式类型是单元模式、左右上下模式、环绕模式还是半环绕模式。然后将字符分解为许多有意义的子结构。每个子结构实际上是一个部首，在四个方向上每个子结构划分为8个子区域，利用每个子区域像素数的统计特征，采用支持向量机进行部首识别。由于汉字是由偏旁组成的，因此汉字的识别问题转化为输入汉字与参考模式之间的一系列偏旁匹配问题。此外，一个字符可能有前偏旁和后偏旁，所以在粗分类阶段利用前偏旁或后偏旁来减少候选字符的数量。由于只需要记录一个字符的偏旁特征和偏旁字符串，大大减少了模板文件的大小，提高了识别速度。另外，由于该系统同时采用了结构特征和统计特征，因此不受笔画数和笔画顺序的影响。
This paper proposes an approach of radical-based online handwritten Chinese character recognition using support vector machine (SVM). In our system, after the input Chinese character is preprocessed, segments ,the structure feature of the character are extracted and the middle point of each segment is projected in horizontal and vertical directions to detect that which type the character's pattern type is of single-element pattern, left-right top-bottom pattern, surrounding or half-surrounding pattern? The character is then decomposed into many meaningful sub-structures. Every substructure is a radical actually and is divided into 8 subareas for each of four directions so that the statistics feature of number of pixels in each subarea is adapted to recognize radical using SVM. Since Chinese character is composed of radicals, the recognition of Chinese character is transformed to series of radical matching between the input character and reference pattern. Furthermore, a character could have front radical and rear radical, so the front or rear radical is utilized in coarse classification stage to reduce the number of candidate characters. In respect that only radicals' feature and the radical string for a character is needed to be recorded, the template file size can be largely reduced, the speed of recognition is improved greatly. In addition, as the structure feature and the statistics feature are both adopted in this system, it is independent on stroke-number and stroke-order.