Skew Detection Using Contour Analysis for a Scanned Journal Page

Image analysis and recognition on a journal or conference paper, called J-image here, is greatly worthy of researching due to existence of various objects and layouts in it. Furthermore, a scanned J-image may be not in a normal orientation, which will usually affect the recognition result. Thus solving the document skew problem is still a challenging topic today. In this paper, a contour analysis based approach is presented to deal with a complex J-image, which may consist of texts, tables, graphics, figures and mathematical expressions with oneor two- or mixed-column page format. A set of 859 skewed J-images with different skew angles within [0, 90] were used for experiments. The mean error of estimating skew angle below 0.1 can be obtained in general cases, which confirms the feasibility of the proposed approach.