Document skew detection using minimum-area bounding rectangle

Abstract

Detection of document skew is an important step in document image analysis. The paper presents a new method for calculation of document skew. The method forms large connected components by a smoothing algorithm and calculates the document skew by finding the orientation of the minimum-area bounding rectangle of one of several connected components. Connection of text to non-text in the smoothing step does not degrade the performance of the method. The smoothing parameters are determined automatically and no manual adjustment is necessary. The method is not limited in the range of detectable skew angles and the achievable accuracy. Experimental results show the high performance of the algorithm in detecting document skew for a variety of documents with different levels of complexity.