A mixture model using Random Rotation Bounding Box to detect table region in document image

Abstract

Table detection in the document image is still a challenging problem due to the variety of table structures and the complexity of document layout. In this paper, we propose a novel method for detecting table regions by using a new shape which is called Random Rotation Bounding Box. This shape is used for illustration and description of the table regions. Based on it, our system performs the following three fundamental steps to detect the table zones: classification of the text and non-text elements in the document image, detection of the ruling-line tables, and identification of the non-ruling-line tables. Different from other methods, our approach can detect most kinds of tables with high precision even when it is skewed. Besides, the proposed method is also designed to fit in the document layout analysis system. Our algorithm has been tested on the two well-known, and a commercial datasets: ICDAR2013 table competition, UNLV, and Diotek. Experimental results on these databases show that our method is more robust and efficient than previous systems.