Skew Estimation in Document Images Based on an Energy Minimization Framework

Skew estimation is important for document analysis and application. Most existing methods are proposed to deal with the document images consisting of words. In most cases, a complex document may include tables, irregular pictures and other non-text components. To address the challenging problem, this paper proposes a novel skew estimation approach based on an energy minimization framework for skewed scanning document images. In the proposed approach, the foreground pixel state information is computed at first. Then a new cost function that considers both background and foreground information for skew estimation is constructed by using state information. A coarse skew is yielded by employing line fitting technique. Then the coarse skew is refined by iteration so that the cost function gets minimum. The ICDAR2013 DISEC dataset is used to evaluate the proposed approach and the experimental results show its effectiveness.