Automatic Orientation Correction of AEC Drawing Documents

摘要

AEC (Architectural, Engineering, Construction) class of drawing documents (see fig. 1, cropped regions) play a pivotal role in the construction and manufacturing domain. These documents comprise of texts and graphical drawings. For various applications, it is necessary to have the text recognized by a computer software. But due to the presence of graphical components (GC) and wrong skew/orientation, it becomes difficult for the OCR system to accurately classify the text regions from drawing documents. To overcome these difficulties, we propose an approach to detect and correct the orientation of the AEC class of drawing documents and also localize the text regions. The proposed work will benefit the OCR system by improving the recognition accuracy. Also, if graphics parts are separated from the text, further processing can be done on graphics alone. Moreover, information of OCRed text and graphical shape can be combined to solve other problems like automatic hyperlinking of callouts [9-10] in AEC documents. This work will also help in the context of content based searching.