This paper presents a novel approach for the detection of tables present in documents,leveraging the potential of deep neural networks. Conventional approaches for table detection rely onheuristics that are error prone and specific to a dataset. In contrast, the presented approach harvests thepotential of data to recognize tables of arbitrary layout. Most of the prior approaches for table detectionare only applicable to PDFs, whereas, the presented approach directly works on images making it generallyapplicable to any format. The presented approach is based on a novel combination of deformable CNNwith faster R-CNN/FPN. Conventional CNN has a fixed receptive field which is problematic for tabledetection since tables can be present at arbitrary scales along with arbitrary transformations (orientation).Deformable convolution conditions its receptive field on the input itself allowing it to mold its receptive fieldaccording to its input. This adaptation of the receptive field enables the network to cater for tables of arbitrarylayout. We evaluated the proposed approach on two major publicly available table detection datasets:ICDAR-2013 and ICDAR-2017 POD. The presented approach was able to surpass the state-of-the-artperformance on both ICDAR-2013 and ICDAR-2017 POD datasets with a F-measure of 0.994 and 0.968,respectively, indicating its effectiveness and superiority for the task of table detection.
@article{pub11283,
author = {
Siddiqui, Shoaib Ahmed
and
Malik, Muhammad Imran
and
Agne, Stefan
and
Dengel, Andreas
and
Ahmed, Sheraz
},
title = {DeCNT: Deep Deformable CNN for Table Detection},
year = {2018},
volume = {6},
pages = {74151--74161},
journal = {IEEE Access},
publisher = {IEEE}
}
Deutsches Forschungszentrum für Künstliche Intelligenz German Research Center for Artificial Intelligence