CAJ | 학술논문

手写体维文识别技术的关键在于特征提取方法的选择。为探求一种可靠性高、可分性强的脱机手写维文特征提取方法，在分析现有方法的基础上，结合维吾尔文字词自身的特点，提出一种在局部特征上基于弹性网格区域笔划密度、方向分解特征，在全局特征上提取交叉点、环、弧形笔划、附加笔划、外围轮廓特征的混合特征提取方法。通过在IFN/ENIT标准数据库及自采样数据集的聚类分析实验，识别正确率分别达到85％、84．3％。结果表明，方法综合统计特征和结构特征提取的优点，具有较强的抗扰能力，可分性优于GS C法。
수사체유문식별기술적관건재우특정제취방법적선택。위탐구일충가고성고、가분성강적탈궤수사유문특정제취방법，재분석현유방법적기출상，결합유오이문자사자신적특점，제출일충재국부특정상기우탄성망격구역필화밀도、방향분해특정，재전국특정상제취교차점、배、호형필화、부가필화、외위륜곽특정적혼합특정제취방법。통과재IFN/ENIT표준수거고급자채양수거집적취류분석실험，식별정학솔분별체도85％、84．3％。결과표명，방법종합통계특정화결구특정제취적우점，구유교강적항우능력，가분성우우GS C법。
The key of recognition technology for offline handwritten Uyghur text is the selection of feature extraction method.In order toexplore a features extraction method for Uyghur text with high reliability and strong separability,based on analysing the existing methods andcombining the characteristics of the Uyghur word itself,we present a mixed features extraction method,which is based on the strokes densitiesin elastic mesh region and the directional feature decomposition in its local features,and extracts the features of the strokes of cross-points,rings and arcs as well as the additional strokes and peripheral contours in its global features.Through the cluster analysis experiments on bothIFN /ENIT benchmark database and self-sampling test set,its recognition correct rate reaches 85% and 84.3% separately.The resultsindicate that this method integrates the advantages of statistical feature and structural feature,has quite strong capability in anti-interference,and its separability is better than GSC method as well.