基于信息融合的维吾尔文联机手写单词识别技术研究

The main difficulty in Uyghur handwritten word recognition is that the basic characters are not precisely segmented, and there are still lots of conjoined sections which can not be directly recognized. This project research an effective approach for online handwritten Uyghur word recognition based on the analysis of the unique shapes and writing styles of Uyghur words. Use of the integration of recognition-segmentation method, the words segment into conjoined sections, and over-segmentation is applied to further segment the conjoined sections into the basic unit of sequences, and merging them to obtain a segmentation candidate grid; the optimal segmentation and recognition result is achieved by fusion of geometric analysis, isolated character classifier and semantic information all-together. The beam search algorithms and dynamic programming algorithm is used for optimal path search in the word recognition process. The geometric information is estimated on current words to adapt to various writing styles of words, it includes unitary and binary geometric information; Recognition information is given by the character classifier with candidate results and their confidence; Semantic information is described by a character based model. The Uyghur language is widely used among the ethnic minorities in Xinjiang and the automatic recognition of handwritten Uyghur characters has huge potential applications in many fields.

维吾尔文手写单词识别中的主要难点是字母在被识别之前不能被精确地切分，仍存在着不能被直接识别的连体段。本项目结合维吾尔文单词的独特手写风格，研究有效的基于分割与识别相集成的方法：利用基于识别的切分方法将单词切分为连体段后，再将其切分成基本单元序列。将连续的基元合并成候选字符并构成切分候选网格；将几何上下文信息、字母识别信息和语言上下文信息一起加入到路径评价准则，得到最优的切分结果及对应的最优识别结果。其中，集束搜索算法和动态规划算法用于单词识别过程中的最优路径搜索。几何信息包括字母一元几何信息和字母间的二元几何信息，是根据当前单词自身的特点统计获得；识别信息由字母分类器给出，包括候选识别结果及其相应的置信度；语义信息用基于字母的语言模型进行描述。维吾尔文字作为一种在新疆少数民族地区和中亚部分地区流行的语言文字，研究其手写文字识别方法对促进少数民族地区的信息化步伐，增进国际交流都是非常有益的。

联机手写输入做为一种自然、方便的输入方法，已经得到了高度重视，并在日常生活中得到了广泛应用。本课题在深入调研国内外联机手写识别技术的研究动态，并通过分析维吾尔文字母与单词自身的结构和书写特点，在联机手写维吾尔文字母识别、维吾尔文单词切分、维吾尔文单词识别等方面做了一些有益的探索。建立了手写维吾尔文字母样本库和手写单词样本库；设计了一个基于DTW的联机手写维吾尔文字母识别系统框架；提出了基于多分类器集成维吾尔文字母识别方法；通过对手写维吾尔文字中的字母连接特点深入研究，提出了一种有效的基于动态规划的联机手写单词分割方案；提出了基于多步分割的维吾尔文单词切分方法；提出了多信息融合的联机手写维吾尔文单词识别框架和方法。维吾尔文字手写输入系统的研究不仅对其他民族文字手写输入系统的研究也有一定的参考价值，而且促进少数民族地区的信息技术发展。