Template Based Table Document Recognition

Abstract

In this paper we proposed an effective but simple table recognition algorithm in the OCR field. First a binary image template is built for the tables to be recognized, which consists of all the lines describing the table cells we are interested in. Then the images to be analyzed are thresholded and deskewed. Vertical and horizontal lines are extracted from the preprocessed image to form a table “scene”. Finally the binary image template is aligned to the table “scene” by minimizing their Hausdorff distance. From the alignment the image regions of interest corresponding to the table cells are extracted for further recognition task. Experiments show that the proposed method with the template can cope with many low quality images and achieve good recognition results.

Improved Faster R-CNN for the Detection Method of Industrial Control Logic Graph Recognition

“… Rho et al (2020) identified the engineering drawings and extracted the row text in the document to automatically generate the building’s 3D model. Jia (2012) used template matching to identify the tables in the drawing to simplify the identification steps.…”