基于多分类器集成的维吾尔文联机手写字母识别

Multi-classifier combination scheme for recognition of online handwritten Uyghur characters

多分类器组合能够在一定程度上弥补单个分类器的缺陷,因此它在模式识别中得到了广泛应用。深入调研国内外联机手写识别技术的研究动态,结合维吾尔文字母的独特书写风格,研究了基于多分类器集成的维吾尔语联机手写字母识别。利用5种不同的特征提取方法构造了5个独立的维吾尔语字母分类识别器,采用了等权投票和不等权投票等两种策略将5种维吾尔语字母分类识别器进行了有效组合。其中,单分类器采用了基于动态时间弯折(DTW)匹配距离的最近邻分类方法。实验结果表明,提出的集成策略的识别率明显高于单分类器的识别率,而且为特征的综合集成提供了多种有效途径。

Combination of multiple classifiers, a certain extent, compensate for defects of a single classifier, so it has been widely applied in pattern recognition. Through in-depth study of the trend of on-line handwriting recognition technology in domestic and abroad, based on analysis of the unique shape of Uyghur characters, this paper studies the Uighur online handwritten character recognition based on multi-classifier combination. It applies five different feature extraction methods to construct five separate classifiers of Uyghuer characters and using voting strategy of ranging from rights to effective combined five kinds of classifier of Uyghuer characters. Each classifier uses the nearest neighbor classification method based on Dynamic Time Warping（DTW） matching distance. Experimental results show that the recognition rate based on integration strategy is significantly higher than the recognition rate of separate classifier, and it also provides a variety of effective ways for the comprehensive integration of features.