CAJ | 학술논문

针对当前计算机辅助注释手写文本文件转录算法存在效率不高的问题，提出一种能预测自动识别单词块中的错误率，并估计校正转录到某个用户定义的错误率所需花费精力的手写文本文件转录算法。首先，分析传统的错误估计方法及其存在的主要问题；然后，提出对整个单词块执行错误估计以提高准确率的思想；最后，将当前执行最好技术进行合并，提出手写文本转录方法。本算法包含在转录手写文本文件的交互式方法中，以主动学习和半监督学习技术有效利用用户交互。在2个真实手写文件上进行转率实验，实验考虑了用户所花精力和转录准确性之间的平衡，实验结果表明了本算法的有效性。
침대당전계산궤보조주석수사문본문건전록산법존재효솔불고적문제，제출일충능예측자동식별단사괴중적착오솔，병고계교정전록도모개용호정의적착오솔소수화비정력적수사문본문건전록산법。수선，분석전통적착오고계방법급기존재적주요문제；연후，제출대정개단사괴집행착오고계이제고준학솔적사상；최후，장당전집행최호기술진행합병，제출수사문본전록방법。본산법포함재전록수사문본문건적교호식방법중，이주동학습화반감독학습기술유효이용용호교호。재2개진실수사문건상진행전솔실험，실험고필료용호소화정력화전록준학성지간적평형，실험결과표명료본산법적유효성。
To solve the problem of poor performance in present computer-assisted annotation transcription of handwritten text doc-uments, a new algorithm for predicting the error rate in a block of automatically recognized words is proposed, and estimates how much effort is required to correct a transcription to a certain user-defined error rate. Firstly, the main problem in traditional error estimating methods is analyzed. Then, the estimation of the error is performed for a whole block of words to raise the accuracy rate. Finally, the best-performing techniques presented in previous works are combined to form our method. The proposed method is included in an interactive approach to transcribe handwritten text documents, which efficiently employs user interactions by means of active and semi-supervised learning techniques. Transcription results, in terms of trade-off between user effort and tran-scription accuracy, are reported for two real handwritten documents, and prove the effectiveness of the proposed algorithm.