Automatic Table Detection, Structure Recognition and Data Extraction from Document Images

Abstract

In the recent advancement, the extensive usage of electronic devices to photograph and upload documents, the requirement for extracting the information present in the unstructured document images is becoming progressively intense. The major obstacle to the objective is, these images often contain information in tabular form and extracting the data from table images presents a series of challenges due to the various layouts and encodings of the tables. It includes the accurate detection of the table present in an image and eventually recognizing the internal structure of the table and extracting the information from it. Although some progress has been made in table detection, obtaining the table contents is still a challenge since this involves more fine-grained table structure (rows and columns) recognition. The digitization of critical information has to be carried out automatically since there are millions of documents. Based on the motivation that AI-based solutions are automating many processors, this work comprises three different stages: First, the table detection using Faster R-CNN algorithm. Second, table internal structure recognition process using morphology operation and refine operation and last the table data extraction using contours algorithm. The dataset used in this work was taken from the UNLV dataset.