Table Information Extraction System (TIES)

Abstract

Table detection and parsing is an actively pursued problem in the domain of document processing. The immense variety in table layouts makes the detection, layout analysis and information extraction from the tables very challenging and complex. The solutions proposed so far fail to generalize on a variety of tables, mainly because they rely on hand-crafted parameters suited for a particular dataset. The accuracy achieved is no better than 60%, due to which none of the solutions can be used for commercial purposes. The utility of Machine Learning in solving this problem has been scarcely explored so far. A data-driven solution is therefore pursued, using ML and Computer Vision algorithms. This approach, after thorough research, can be used to tackle the multitude of challenges involved in table detection and parsing. The pursued solution is aimed to be generic and strives at promising a higher level of accuracy than current systems.