隨著數位學習的普及，越來越多的用戶在智能手機或平板電腦上使用智能筆書寫筆記，手寫辨識技術變得越來越重要，且對於母語較多元的國家來說，多語種手寫文本的數量也越來越多。目前許多辨識系統具有很高的準確度。例如： Google, My-Script 等等，他們在處理手寫文本的辨識這方面已經做得非常地足夠，可是若是讓他們辨識多語種的文檔，其效果就不是原先那麼的好。 為了能夠正確識別這些多語種的手寫文檔，語種識別非常重要。基於國中小學的中文數學應用題之答題文本，本文提出了一個使用計算機視覺、卷積神經網絡以及數學結構表示演算法進行輕量級語種識別系統，對中文字區塊和數學式區塊進行分類，以對應強大的辨識引擎，提高數學應用問題中之作答文本內容的準確性。實驗結果表明，增加了語種識別能夠改善辨識系統，提升了手寫文本識別的正確率。;With the popularity of digital learning, more and more users are using smart pens to write notes on their smartphones or tablets. Handwriting recognition technology is becoming more and more important, and for multi-language countries, the number of multi-script/multi-language handwritten documents is also increasing. In order to correctly recognize these multi-script/multi-language handwritten documents, script identification is very important. Based on the answer documents of Chinese mathematical application for elementary and national schools, this paper proposes a lightweight script identification using computer vision, Convolutional Neural Networks (CNN) and Mathematical Expressions (ME) algorithms to classify Chinese characters and mathematical areas. These areas can be mapped to a powerful recognition engine to improve the accuracy of the answer documents of Chinese mathematical application. The experimental results show that the addition of script identification can improve the recognition system and improve the accuracy of handwritten document recognition.
