工程圖影像字元萃取實作

Implementation of Character Extraction for Engineering Drawing Images

早期工程圖影像以紙本方式來保存，需利用掃描儀器數位化建檔儲存電腦後，可以進一步在工程圖影像開發相關的應用。本文將實作工程圖影像中的字元萃取。工程圖影像經過二值化處理可把背景與物件分離，依此影像中所含特徵，利用串接應用進行斷裂線段之填補，並找出影像中線段與文字的所有可能群聚區域，再以形態學影像處理的運算來重建影像。最後，利用連通物件標記法以及一些判斷條件來分離出字元與線段來提高萃取的正確率。實驗結果顯示本文所用的方法可以成功萃取出工程圖影像字元的位置，獲得相當高的萃取正確率。 Engineering drawing images are saved in paper form initially. We can then scan those images and save them on a computer if we wish to subsequently develop some of their applications. This paper explores the implementation of character extraction for engineering drawings. An engineering drawing image is binarized to separate the background from the object. For engineering drawings, we use cascading applications to fill lines and use connections of points to group lines and writing regions. Then, two forms of morphological image processing, namely dilation and erosion, are applied to rebuild it. Extraction errors result when characters and lines are close together. To solve this problem, connected component labeling and judgment conditions are used to separate characters and lines, thus improving extraction accuracy. Experimental results demonstrate that this method can be used to identify character locations for engineering drawing images, which produces higher quality simply structured engineering drawing images.