基于线条识别的扫描工程图字线分割方法

在扫描工程图矢量化中，文字位图与图形位图分割的问题一直被广泛关注，许多现有方法根据连通区的形态特征和空间分布来分割文字和图形的位图，然后分别识别，其处理能力都局限在无粘连情况下，无法处理字线粘连及相交的情况，且难以区分尺寸范围相近或形态相似的文字与图形，提出一种基于线条识别的文字／图形分割方法，通过准确地识别线条并抽取其位图来分离文字，不仅保证了分割的正确性，而且为文字位图的字串的组织给出了依据。

The Segmentation of Text from Line Based on the Line Recognition in Scanned Engineering Drawings

The problem of text/graphics segmentation has received a great deal of attention in the research of engineering drawing vectorization. Many existing methods perform the segmentation based on the spatial distribution analysis of connected components, which separates the connected components into graphics and texts before the recognition. They are capable of segmenting isolated texts, with the limitation that they do not handle text touching graphics well. In this paper, a recognition based segmentation method is proposed. It segments texts by recognizing lines in their entirety and then extracting the bitmaps corresponding only to the recognized lines accurately. Some verification mechanisms are used to prevent the line like strokes of characters from being misrecognized as lines. Keeping the high accuracy of segmenting isolated texts, the proposed method yields a better result of text touching graphics than existing methods do. Moreover, the recognized lines become the reference to string direction when grouping characters into text strings. Finally, the experimental results of recognition based segmentation using real life engineering drawings is presented.