CAJ | 학술논문

提出了一种基于小波统计特征的朝汉文种识别方法。首先计算行文档图像的垂直及水平方向的投影；其次对垂直方向的投影及水平方向的投影进行一维小波分解并分别计算小波统计特征，然后将两个方向上的小波统计特征进行合并作为该文档图像的特征向量（本文方法构造的特征向量仅为11维）；最后通过神经网络进行训练和测试，结果显示平均测试准确率超过94％。
제출료일충기우소파통계특정적조한문충식별방법。수선계산행문당도상적수직급수평방향적투영；기차대수직방향적투영급수평방향적투영진행일유소파분해병분별계산소파통계특정，연후장량개방향상적소파통계특정진행합병작위해문당도상적특정향량（본문방법구조적특정향량부위11유）；최후통과신경망락진행훈련화측시，결과현시평균측시준학솔초과94％。
A script identification method between Chinese and Korean language based on wavelet statistic fea-ture is presented.To reduce the dimension and improve calculation efficiency,each 2D row-document image partitioned from original document image is converted into 1D projection signal in both vertical and horizontal direction.1D wavelet decomposition is implemented on the both projections.Then,wavelet statistic features are calculated for both proj ections and merged as feature vector of row-document.Effectiveness of wavelet sta-tistic feature vector is evaluated by BP neural network.The experimental results show that the identification accuracy average around 94%.